Ecostat

#AIFORALL - Compendium of Project Reports prepared by the officers in Department of Economics & Statistics who have attended the 45 days programme "Capacity Building in Artificial Intelligence & Data Analytics" conducted by Digital University Kerala from 28-11-2022 to 31-01-2023.
Published Date: 30, May 2023
Jurisdiction : Kerala
Scheme : Research Papers
View Count : 2431
Download
Related Publications
Compendium of Project Reports 2024
Published Date : 27, Feb 2024
Compendium of Project Reports prepared by DUl #AIFORALL batch 2 from Department of Economics & statistics from 1-10-2023 to 21-12-2023
Ecostat Bulletin March 2024
Published Date : 15, Feb 2024
Ecostat Bulletin January 2024- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of February 2024
Ecostat Bulletin February 2024
Published Date : 10, Feb 2024
Ecostat Bulletin February 2024 -Programmes and events conducted by the Department of Economics & Statistics, SASA and Kerala State Statistical Commission during January 2024
Ecostat Bulletin December 2023
Published Date : 27, Jan 2024
Ecostat Bulletin December 2023 Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of November 2023
Ecostat Bulletin January 2024
Published Date : 27, Jan 2024
Ecostat Bulletin January 2024- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of December 2023
Ecostat News May 2023
Published Date : 02, May 2023
Ecostat News Magazine for the month of May 2023 - the official magazine of Department of Economics & Statistics
Ecostat News January 2023
Published Date : 01, Jan 2023
Ecostat News Magazine for the month of January 2023 - official magazine of Department of Economics & Statistics
Ecostat News November 2022
Published Date : 01, Nov 2022
Ecostat News Magazine November 2022
Administration Report of Department of Economics & Statistics 2018-19
Published Date : 22, Apr 2022
Administration Report of Department of Economics & Statistics 2018-19
Decentralized Planning Survey 2015-16 - 2019-20
Published Date : 10, Nov 2021
Decentralized Planning Survey 2015-16 - 2019-20
MORE PUBLICATIONS
Transcript

Government of Kerala 	
 
 
 
  
 	
COMPENDIUM OF PROJECT REPORTS 	
 
 
 
 	
#AIFORALL 	
Capacity Building in  	
Artificial Intelligence & Data Analytics  	 	
 
 	
Conducted by 	
 	
Kerala University of Digital Sciences Innovation and Technology 	
 	
From 28	th	 Nov 2022 to 31	st Jan 2023 	
 	
In collaboration with 	
  	
Department of Electronics & Information Technology Kerala 	
& 	
Kerala State Information Technology Mission  	
 	
May 2023 	
 	
DIRECTORATE OF ECONOMICS & STATISTICS 	
VIKAS BHAVAN, THIRUVANANTHAPURAM

PUNEET KUMAR	  IAS	
ADDITIONAL  CHIEF  SECRETARY	
 Government of Kerala	
Planning & Economic Affaris, 	PIE & M Dept.	
Government  Secretariat	
MESSAGE	
     T h i r u v a n a n t h a p u r a m- 6 9 5  0 0 1	
      Phone : O471-2335452, 0471-2518356	
           	     Mobile : +91-9910816200	
       E - m a i l  :  p rl s e cy. p l g@ k er a l a. go v. i n	
      Date : 19/05/2023	
It is learnt that a capacity building programme on Artificial Intelligence and Data Analytics has been	
conducted by Kerala University of Digital Sciences Innovation and Technology for statistical staff of the
Department of Economics and Statistics under Planning & Economic Affairs Department. This training
course was organized with the budgetary support of Department of Electronics and Information Technol-
ogy Kerala and Kerala State Information Technology Mission. This 9 weeks course was exclusively con-
ducted for statistical personnel in the cadre of Joint Director to Statistical Assistant working in Directorate
and other departments. This course includes advanced learning topics- Machine Learning techniques like
Regression Analysis. Neural Networks, Artificial Intelligence, Statistical Prediction etc. This Compendium
contains the reports of projects done by each trainee as part of the capacity building programme by
analyzing the datasets of various subjects which are available in the department and other socio economic
sectors of the State.	
I think this advanced learning programme on Data Science will be much useful to a department like	
Economics and Statistics, the Nodal Agency of official statistics in the State. This will lead to encourage the
use of latest data science tools with the help of technology for analyzing and disseminating better results to
the planners. In this occasion I thank Vice Chancellor Digital University Kerala, Director of Economics and
Statistics and other officials in DUK and DES, who were behind the conduct of this programme very
effectively.	
I hope this Compendium of Project Reports will be a worthy reference for the statistical staff in	
DES, planners, academic institutions, research agencies, research aspirants and those working in the field
of data science.	
(PUNEET KUMAR IAS)

MESSAGE	
Capacity building initiatives are of prime importance to develop in-house IT teams that	
are fully equipped to deal with the complexities of managing and   implementing IT projects and to
reduce         dependency on external agencies. As an organisation like  Department of Economics &
Statistics which is dealing with official statistics, capacity building in Data Science is essential to equip
the statistical   personnel to the level of data scientists and to utilise them to build quality research
outputs to cater the needs of Government planning purpose.	
State IT department conducts Capacity Building programme in Artificial  Intelligence and Data	
Analytics through Digital University Kerala with the objective of equipping a team of officers in
Government system to understand the latest data  science tools and utilise the technology in effective
decision making. I am happy to learn that the Department of Economics & Statistics has participated
the second batch of this advanced programme and a team of 19 officials from DES has  successfully
completed this course during the year 2022-23.	
I hope that this learning initiative will lead to build a team of statistical personnel to conduct	
research in official statistics and thus to empower the Statistical System in Kerala. I congratulate the
Vice Chancellor DUK and his team for arranging the second batch of  Al & DA  exclusively for DES
staff and appreciate the effort taken by the Director and his team in DES for participating in it.
Dr. RATHAN  U. KELKAR  	IAS                            	Electronics & Information Technology	
   	 SECRETARY TO GOVERNMENT	           	Taxes and Excise Department
     	       Government Secretariat	
         Thiruvananthapuram, Govt. of Kerala	
  E - m ai l  :  s e cy. i t d @ ke r a l a . gov. i n	
 s e c y. t a x e s @k e r a l a . go v. i n	
 Government of Kerala	
Phone 	{	:0471-2336602	
Office : 0471-2518444	
Date 18-05-2023	
Dr. RATHAN U. KELKAR IAS

MESSAGE	
In order to realize the Keralaâs commitment to become a truly digital State where advanced	
technologies are used by Government for improving citizen services, Digital University Kerala (DUK) has
started programs on imparting high end technology skills to Government employees in a structured manner.
I am happy that twenty officers from the Department of Economics and Statistics (DES) with proven
expertise in Statistics, have been trained in Artificial Intelligence and Data Analytics at DUK, as part of this
initiative being implemented with the funding support from the Department of Electronics and IT, Government
of Kerala.  DUKâs training and technology development services deliver skill development among
Government and professionals to prepare them to lead the digital transformation of the future. DUK
capacity building training nurtures them to build their own services to address specific challenges faced by
the Department and build systems using cutting edge technologies like AI and data analytics.  Fostering a
knowledge economy, DUK continues to expand its offering to new courses and technologies which can
match government and industries with a world class digital education experience.	
As DES is the state nodal agency for collecting and disseminating data on various socio-economic	
sectors of Kerala, it is indeed a great pleasure that DES has taken an innovative step to leverage the
technologies for building such services to transform the activities of DES to enhance its role in official
statistics.  Hope the participants of course could translate their newly acquired skills to improve your
services and making better decision making with increased transparency and intelligence.
Wish you all the Best.
Prof. Saji Gopinath	
 	
Kerala University of Digital Sciences	
    Innovation & Technology	
Te c h n o c i t y  C a m p u s ,  M a n g a l a p u r a m	
Thiruvananthapuram- 695317	
Date: 25-05-2023	
Prof (Dr.) Saji Gopinath	
Vi c e  C h a n c e l l o r

Kerala University of Digital Sciences	
        Innovation & Technology	
Te c h n o c i t y  C a m p u s ,  M a n g a l a p u r a m	
Thiruvananthapuram- 695317	
Date: 25-05-2023	
ACKNOWLEDGEMENT	
Recognizing AIâs potential to transform economics of Kerala, Digital University Kerala (DUK)âs	
pursuance in upskilling Government employees for current and future technologies position themselves as
leaders with unique brand of  #AIFORALL.   #AIFORDES aims to empower human capabilities to
enhance DES skilled expertise for effective implementation of AI to evolve scalable solutions for emerging
economics	.  Th e  ca p ac i t y b u i l d i n g i m p ar t e d t o  D E S  w a s  a  gr e at  s uc c es s  wi t h t h e e f fo r t  o f  D E S  o f fi c i al s	
and DUK.  A 45 days intensive customized training with hands-own experience using DES data for their
services and applications enriched the programme with greater success.  Remarkable contributions have
been made by the participants by implementing AI projects in critical areas such as health, Agriculture,
Consumer Price indexing etc using Machine Learning and Deep learning techniques. Forecasting of consumer
price indexing in Kerala, Predictions on CRS Data â Live birth, Death, Maternal Death and Infant Death,
Prediction of Cost of Cultivation of Important Crops, paddy cultivations in Kerala etc. are a few
Congratulations to all the participants who have successfully completed the course and projects and we
are grateful to the Director Mr. Sajeevu P P and Deputy Director Mr. D S Shibukumar for constant
interaction and cooperation for the training.  The programme would not have been a success without the
dedication and sincere efforts of trainers from Virtual Resource Centre for Language Technology (VRCLC)
of DUK and External trainers Mr. Prasad K Nair, Project Head, GBS Technologies Trivandrum, Mr.
Somasekharan, Professor (Rtd) in Statistics, University College  and Dr. Satheesh Kumar, Professor,
Futures Studies of University of Kerala.
I am grateful to Dr. Saji Gopinath, Vice Chancellor of Digital University Kerala for his instinct support and
encouragement throughout the training. We are indebted IT Department, Government of Kerala for their
vision to enhance the capabilities of Government employees to meet the futuristic technologies for their
services and the financial support to conduct the capacity building.
 E l i z a b e t h  S h e rl y	
 	
Prof. Elizabeth Sherly
Distinguished Professor	
sherly@duk.ac.in

Government of Kerala	
PREFACE	
Strengthening the statistical system in Kerala is the vision of the Department of Economics and	
Statistics and in line with this capacity building programmes and training programmes are being conducted
every year through State Academy on Statistical Administration and Institute of Management in Government
Kerala.	
This project reports are prepared by statistical personnel as part of the capacity building programme	
on Artificial Intelligence and Data Analytics which was conducted by Digital University Kerala for 45 days
from 28 November 2022 to 31 January 2023. This course was offered by DUK and sought nominations
from DES. DES has sent nominations of 20 statistical staff and requested DUK to consider this programme
as an exclusive batch of DES staff. DUK has agreed the request of DES and accordingly this adva learning
programme was conducted at IIITM-K, Technopark for the convenience of DES.	
This course was much useful for the staff in DES as the department is dealing in official statistics	
and several surveys and studies are being conducted in various socio-economic subjects under the guidance
and technical support of Govt of India. In this occasion, I would like to express my sincere gratitude to
DUK especially Dr. Saji Gopinath, Vice Chancellor DUK, Dr. Elizabeth Sherly, Programme Coordinator
DUK and Sri. Surag M, Assistant Coordinator DUK for arranging the programme as an exclusive batch
of DES staff and for their sincere and timely intervention on all matters for the successful completion of the
course in an effective manner.	
I also express my sincere thanks to Additional Director Smt. Lathakumari C.S. and Deputy Director	
Sri. D.S. Shibukumar for their painstaking efforts for the successful accomplishment of this two months
programme in a very fruitful manner.	
I would like to appreciate the officers who have participated and taken effort for completion of this	
capacity building initiative. I also express my thanks to the controlling officers who have given permission
to attend this 45 days course without affecting the normal duties.	
I hope that this compendium of reports will be of great useful for the department, research aspirants,	
academicians, colleges and universities and all those who are interested in data analytics and predictive
learning in official statistics. Suggestions for improvement of the programme and contents of this report are
most welcome.	
Sajeevu P.P.	
Department of Economics & Statistics	
Vi ka s Bh av an  ,  T hi ru va na ntha pu ra m- 6 95 03 3	
Phone: 0471- 2305318, Fax: 0471- 2305317	
Email: ecostatdir@gmail.com	
Date: 25-05-2023	
SAJEEVU P.P.
  DIRECTOR

Trainees of the Programme
1 Smt. Sudarsha R., Joint Director, Survey & Design Division
2 Smt. Shailamma K., Deputy Director, Evaluation Division
3 Sri. Preeth V.S., Deputy Director, MCCD Division
4 Smt. Maya R., Deputy Director, O/o Chief Town Planner
5 Smt. Dhanya A., Deputy Director, O/o IDRB
6 Sri. Vijay R., Research Officer, Survey & Design Division
7 Sri. Sijith K.S., Research Officer, Labour & Housing Division
8 Sri. Abhilash K.V., Research Officer, ASI Division
9 Smt. Praseeda Gopan, Research Assistant, State Income
10 Smt. Suma S.A., Research Assistant, Evaluation Division
11 Sri. Prasanth B.R., Computer Supervisor, Computer Division
12 Sri. Rajesh R., Statistical Assistant Grade I, Computer Division
13 Sri. Ashad W.A., Statistical Assistant Grade I, Computer Division
14 Smt. Neethymol Kurian, Statistical Assistant Grade II, BSLLD Division
15 Smt. Chitra J.V., Statistical Assistant Grade II, District Office, Thiruvananthapuram
16 Smt. Brijila A.S., Statistical Assistant Grade II, District Office, Thiruvananthapuram
17 Kumari. Minu Merin Andrews, Statistical Assistant Grade II, PPC Division
18 Kumari. Baby Sindhu, Statistical Assistant Grade II, ASI Division
19 Sri.Nidhin Babu M., Statistical Assistant Grade II, District Office,
               Thiruvananthapuram	
I

BEHIND THE PROGRAMME
Coordinators
Department of Economics & Statistics	
1. Smt. Lathakumari C.S., Additional Director (General)
2. Sri. D.S. Shibukumar., Deputy Director, Computer Division	
Digital University Kerala	
1. Dr. Elizabeth Sherly, Distinguished Faculty DUK
2. Sri. Surag M., Scientific Associate (Training & Development) DUK	
Faculty  Members
Digital University Kerala	
3. Dr. Elizabeth Sherly, Distinguished Faculty DUK
4. Dr. Malu G., Research officer, DUK
5. Sri. Surag M., Scientific Associate (Training & Development) DUK
6. Smt. Nayana Uday, Research Scholar, DUK
7. Smt. Leena G. Pillai, Scientist, DUK
8. Smt. Judy K. George, Research Scholar, DUK
9. Smt. Sabitha Rani B. S., Research Scholar, DUK	
Other Organizations	
1. Sri. Prasad K. Nair, Project Manager, GBS Technologies Trivandrum
2. Dr. Satheeshkumar Krishnan Nair, Professor, Department of Futures Studies, University of	
Kerala	
3. Smt. Gayatri Menon, Research Scholar, Kerala University
4. Sri. Somasekharan Pillai, Former HoD, Department of Statistics, University College,	
Thiruvananthapuram	
5. Smt. Raji Gopinath, Information Officer, Entrance Commissionerate, Thiruvananthapuram	
II

INTRODUCTION	
The Department of Economics and Statistics is the Nodal Agency in Kerala for collection, compilation	
and analysis and dissemination of data on all socio-economic sectors of the State economy. Kerala has a
well-established statistical system in India to organize and conduct sample surveys and census for the
planning purpose of State and Central Governments. The department is operationalizing many schemes as
per the guidelines and methodological support of Central Statistical Office and National Sample Survey
Office under National Statistical Office of the Ministry of Statistics and Programme Implementation
(MOSPI), Government of India. The details of schemes of surveys conducted by the department are
mentioned in the official website 	https://www.ecostat.kerala.gov.in	.	
The department is publishing on an average 30 reports of various categories in each year and the	
digital copy in pdf format are available in official website. So far, the department has published more than
500 reports online. The digital publication started from 2004. In addition about 900 reports, which were
published earlier, are available in the department library in printed format.	
Conventional methods of statistical analysis are usually performed in the reports of the department.	
In this modern world, data is the oil for planning and Data Science provides latest techniques by blending
the ICT technology for analysis the data more efficiently and to provide a best result for effective decision
making in a quick and easier manner. Data science is the study of data to extract meaningful insights for
business. It is a multidisciplinary approach that combines principles and practices from the fields of
mathematics, statistics, artificial intelligence, and computer engineering to analyze large amounts of data.	
Data Science combines Mathematics and Statistics, Specialized Programming, Advanced Analytics,	
Artificial Intelligence (AI), and Machine Learning with specific subject matter expertise to uncover actionable
insights hidden in an organizationâs data. These insights can be used to guide decision making and strategic
planning. Data scientists rely on popular programming languages to conduct exploratory data analysis and
statistical regression. These open source tools support pre-built statistical modeling, machine learning, and
graphics capabilities. Python and R are the most popular programming languages in Data Science.	
This Compendium consists of Project Reports prepared and submitted by a group of statistical	
personnel of the Department of Economics and Statistics as part of the Capacity Building Programme on
Artificial Intelligence and Data Analytics which was conducted by Kerala University of Digital Sciences
Innovation and Technology, which is popularly known as Digital University Kerala (DUK) during the year
2022-23. This programme of AI & DA is the second batch of the DUK with the funding of State IT
department and Kerala State IT Mission. The programme was conducted in 45 working days, started
from 28 November 2022 and ended on 31 January 2023. For the convenience of the department and
offices where the trainees are working, considering the long duration of the course, the programme was
scheduled from 9.00 AM to 1.00 PM. The faculty support, lunch and light refreshment charges were
borne by DUK. The 45 days course was successfully completed on 31 January 2023 with project
presentation by each participant. After the completion of the programme an orientation programme on this
subject has been conducted by DES for the directorate staff in February 2023. Project presentations were
arranged before the directorate staff during the months of March and April 2023 for understanding the
data  analysis  done  by  the  trainees  and  to  give  a  motivation  to  the  rest  of  the  staff.
Evolution of the programme	
DUK has called for nominations form DES for joining the programme of second batch AI & DA	
conducted by DUK with the budgetary support of Govt of Kerala. DES has sought willingness from the
directorate staff. Willingness was sought from some staff working in line departments and district office of
Thiruvananthapuram. Since the programme, as informed by DUK, was to be conducted on a part time
manner with FN at DUK and AN at concerned office for normal office duties, priority was given to offices
in Thiruvananthapuram headquarters. Priority was also given to statistical personnel having MSc degree in
Statistics or Mathematics as the data analysts must know some statistical theories and techniques to
understand the outcome. More than 20 nominations were obtained, after screening 20 officials were
III

selected and informed to DUK and requested DUK to consider this as an exclusive batch for DES staff.
DUK has agreed with the suggestion of DES and then proceeded to implement.
Digital  University  Kerala	
DUK is a premier institution of excellence in science, technology and management. It actively promotes	
higher education through its IT facilitated education programs and services across Kerala and beyond.
The institution is well-known for its research in Artificial Intelligence, Computational Linguistics and Remote
Sensing among others. The institute is a pioneer in conceptualizing and implementing some of Indiaâs well
recognized IT initiatives in education, agriculture and e-Governance.
Programme syllabus- an overview	
Introduction to Python, Python Data Types, control statements in Python, Python functions, modules	
and packages, Python string, list and dictionary manipulation, Python file operation, Python Data Science-
NumPy, MatPlotLib, SciPy, Introduction to machine learning, Regression, clustering , Attribute Selection,
classification, SVM, Neural Networks, Text processing. Introduction to Pandas, SeaBorn,   Exploratory
Data Analysis (EDA) for data, Data cleaning (Null Value Analysis and removal, Dimensionality Reduction,
Mean, Median, Mode, Quartile, Correlation analysis, Outlier Detection and removal), Data Pre-processing,
Examples of Predictions using Various Models. Introduction to R, RStudio and data analysis in R.
Project Assignment and Evaluation	
Tw o  w e e k s  p r o j e c t  a s s i g n m e n t s  w e r e  g i v e n  t o  a l l  t r a i n e e s  d u r i n g  t h e  p r o g r a m m e .  Tr a i n e e s  h a v e	
selected their own subjects and datasets for analysis and project preparation. Majority of the trainees used
DES datasets for data analysis and project preparation.
Datasets used for project preparation	
Crop statistics published by DES under EARAS scheme, daily retail prices of essential commodities	
collected and published by DES, data on Indices of Industrial Production, industrial data under the scheme
Annual Survey of Industries, cost of cultivation data under Cost of Cultivation Survey conducted by DES,
Consumer Price Indices published by DES, Building construction cost data collected by DES quarterly
from selected centers in the State, land utilization data under EARAS scheme, Birth and death statistics
collected under Civil registration System, cause of death data under Medical Certification of Cause of
Death (MCCD) scheme are the data used by the participants for project preparation.
Methods used for analysis	
Machine Learning Models- Long Short Term Memory (LSTM) models, Auto Regressive Integrated	
Moving Average (ARIMA), Artificial Neural Networks (ANN), Support Vector Machines (SVM), Random
Forest etc. are the methods used by the trainees for analyzing and forecasting data.
Project  Reports	
This compendium contains project reports submitted by the participants. In addition, some of the	
trainees have built user interface in Python for inputting data for applying in the model found by them and
retrieving results dynamically.
Future perspective	
This course on AI & DA was extremely useful for the staff in an organization like Department of	
Economics & Statistics which is dealing with official statistics. Definitely this kind of initiatives will lead to
set up a Data Analytics Unit in the department in the nearing future. Data Analytics Wing is essential to think
like a data scientist and analyses the data and converting the department functions into the style of research
oriented activities. The department has a lot of datasets in official statistics and the scope for exploratory
data analysis with the support of computer science and technology and preparing beautiful reports will be
an essential need in the modern world of Data Science. DES is one of the prime departments as data is
essential for planning; a transition from conventional style of data analysis is expected from DES by the
planners community. Such a transition is inevitable.
IV

INDEX OF PROJECT REPORTS 	
 	
Sl No  Project Title  Page No 	
1 	
Machine learning models for predicting rice and tapioca yields in 
Kerala - A comparative analysis 	
Smt. Chitra J.V. & Sri. Ashad W.A.	 	
1 	
2 	
Prediction of cost of cultivation of important crops in Kerala- 
using Machine Learning 	
Smt. Shailamma K.	 	
13 	
3 	
Forecasting cement prices in Kerala 	
Sri. Sijith K.S.	 	
29 	
4 	
Index of Industrial Production- An overview 	
Smt. Sudarsha R. & Smt. Brijila A.S. 	 	
39 	
5 	
Prediction of cost of cultivation of important crops in Kerala- 
using Machine Learning (second article) 	
Smt. Suma S. A.	 	
49 	
6 	
Utilizing AI for accurate forecasting of Keralaâs GSVA and 
NSVA- A methodological analysis and evaluation of benefits 	
Smt. Praseeda Gopan	 	
65 	
7 	
Analysis and predicting the market price of consumer products 	
Sri. Nidhin Babu	 	
71 	
8 	
Crop yield prediction using random forest algorithm for different 
crops in Kerala 	
Smt. Neethymol Kurian	 	
79 	
9 	
Changing trend in cause of deaths in Kerala- An analysis of cause 
of death over the years from 2012 to 2021 	
Kum. Minu Merin Andrews	 	
85 	
10 	
Time series analysis of GSVA of organized manufacturing sector 
of India and Kerala 	
Kum. Baby Sindhu	 	
95 	
11 	
Prediction	 	on live births	 	
Sri. Prasanth B.R.	 	
117 	
12 	
Analyzing the correlation between birth and maternal factors - A 
review of CRS data using Machine Learning approach 	
Sri. Rajesh R.	 	
127 	
13 	
A machine learning approach to classification of cause of death 	
Sri. Preeth V.S.	 	
135 	
14 	
Exploration of land utilization in Kerala- Forecasting 	
Sri. Vijay R. 	 	
147 	
15 	
Building construction cost and index prediction in Kerala- A 
machine learning approach 	
Smt. Dhanya A.	 	
157 	
16 	
To predict daily prices of essential commodities 	
Sri. Abhilash K.V.	 	
165 	
17 	
Forecasting of Consumer Price Index in Kerala- A machine 
learning approach 	
Smt. Maya R.	 	
173 	
 	
V

1 	
Compendium  of  Project  Reports 	
 	 	
Machine Learning Models for predicting 	
Rice and Tapioca yield in Kerala	-	 	 	
A comparative analysis	
 	
Submitted By 	
 	
Smt. Chithra J.V., 	
Statistical Assistant Grade II and	
 	
Sri. Ashad W.A., 	
Statistical Assistant Grade I

2                  	Compendium  of  Project  Reports	 	
 
1.	 	Introduction 
 
Analyzing  data  on  the  production,  consumption,  and  trends  of  Kerala  paddy  and 	
tapioca  can  provide  insights  into  the  agricultural  sector's  overall  health  and  potential 
economic  impact.  Here  are  some  potential  areas  of  relevance  for  analyzing  data  on  these 
crops: 	
ï	 	Agricultural  policies  and  planning:  Data  on  production  and  consumption  can 
inform  policymakers  about  the  current  status  of  paddy  and  tapioca  farming  in 
Kerala.  This  information  can  help  them  develop  effective  policies  and  plans  for 
improving the agricultural sector, increasing crop yields, and supporting farmers. 	
ï	 	Market trends and opportunities: Analyzing data on prices, demand, and supply can 
help  farmers  and  agricultural  businesses  identify  market  trends  and  opportunities 
for  expanding  their  operations.  They  can  use  this  information  to  make  informed 
decisions about crop selection, marketing strategies, and pricing. 	
ï	 	Resource  management:  Data  on  crop  yields,  land  use,  and  farming  practices  can 
help  farmers  make  informed  decisions  about  resource  management.  For  example, 
they  can  use  this  information  to  determine  the  most  efficient  and  sustainable 
methods for irrigation, fertilization, and pest control. 	
ï	 	Climate  change  adaptation:  As  climate  change  continues  to  affect  the  agricultural 
sector,  analyzing  data  on  crop  production  can  help  farmers  and  policymakers 
understand  the  impact  of  changing  weather  patterns  on  crop  yields.  They  can  use 
this  information  to  develop  strategies  for  adapting  to  these  changes,  such  as 
implementing  new  farming  practices  or  crop  varieties  that  are  more  resilient  to 
extreme weather conditions. 	
Overall,  analyzing  data  on  Kerala  paddy  and  tapioca  production  and  consumption  can 
provide  valuable  insights  into  the  agricultural  sector's  health  and  potential  economic 
impact.  This  information  can  help  policymakers,  farmers,  and  businesses  make  informed 
decisions  about  resource  management,  market  opportunities,  and  adaptation  to  climate 
change. 
                                                        
2.	 	Objective 
 
 India's agricultural yield is heavily dependent on weather conditions, especially for 	
rice  cultivation  that  heavily  relies  on  rainfall.  This  makes  timely  predictions  and  analyses 
of  future  crop  productivity  critical  to  help  farmers  maximize  their  yield.  Unfortunately, 
predicting crop yield has been a significant agricultural challenge, as farmers typically rely 
on  past  experiences  with  yield  to  predict  future  harvests.  To  address  this  issue,  various 
techniques  and  algorithms  have  been  developed  for  crop  yield  prediction.  The  role  of  Big 
data  analytics  techniques  in  agriculture  has  also  contributed  to  the  effectiveness  of  these 
techniques. 
Moreover,  forecasting  the  yield  of  paddy  and  Tapioca  is  essential  for  planning  purposes, 
and rice import policies should be based on such predictions. This study aims to investigate 
the  past,  present,  and  future  trends  of  paddy  and  Tapioca  production  in  Kerala.  There  are 
multiple  ways  to  increase  and  improve  the  crop  yield  and  the  quality  of  the  crops.  Data 
mining is also useful for predicting crop yield production.

3 	
Compendium  of  Project  Reports 
 The main objectives are   	
ï¶	 To Analyze the trends in area, production and productivity of Paddy and Tapioca in 
Kerala. 	
ï¶	 To use machine learning techniques to predict crop yield.  	
ï¶	 To analyze climatic parameter (Rainfall)   	
 
3.	 	Methodology and Methods used 	 	
 
The  proposed  model  makes  use  of  two  well-known  machine  learning  algorithms 	
applied to the customized input datasets for rice and Tapioca Separately. They are 	
 	
3.1	 	Linear Regression  	
Linear  Regression establishes  a relationship  between dependent  variable  (Y) and 	
one  or  more independent variables  (X) using  a best  fit  straight  line (also  known  as 
regression  line).  It  is represented  by  an  equation Y=a+bX  +  e,  where  a  is  intercept,  b  is 
slope of the line and e is error term. This equation can be used to predict the value of target 
variable based on given predictor variable(s). 	
The  difference  between  simple  linear  regression  and  multiple  linear  regression  is 	
that,  multiple  linear  regression  has  (>1)  independent  variables,  whereas  simple  linear 
regression has only 1 independent variable. 
Important Points: 	
ï· 	There must be 	linear relationship	 between independent and dependent variables 	
ï· 	Multiple  regression  suffers  from 	multicollinearity,  autocorrelation, 	
heteroscedasticity	. 	
ï· 	Linear Regression is very  sensitive to 	Outliers	.  It can terribly affect the  regression 	
line and eventually the forecasted values. 	
ï· 	Multicollinearity  can  increase  the  variance  of  the  coefficient  estimates  and  make 
the  estimates  very  sensitive  to  minor  changes  in  the  model.  The  result  is  that  the 
coefficient estimates are unstable.  	
3.2	 	Linear Support Vector Regression	 	
 
It is  the generalization of support vector machine (SVM) for regression problems. 	
SVR  can  be  linear  or  non-linear  based  on  the  kernel  function.  In  case  of  regression 
problem,  in  place of  finding  hyperplane,  SVR  find  epsilon  in  sensitive region  around  the 
function that have at most epsilon-deviation, known as epsilon tube.  	
The  models  were  trained  on  the  training  set  and  evaluated  using  the  cross 	
validation technique to predict the crop yield.

4                  	Compendium  of  Project  Reports	 	
 
4.	 	Dataset Used 
 
The dataset used for this project consists of Season-wise data on Rice and Tapioca, 	
including  information  on  Area,  Rainfall,  and  Production  of  14  districts  in Kerala  from  the 
year 2005 to 2020. The dataset contains 1260 rows and 5 columns, including data on Area  
in hectares, Production in tons, and Rainfall in millimeters. Both numerical and categorical 
attributes  are  included  in  the  dataset.  The  data  were  collected  from  the  Agricultural 
Statistics  of  the  Department  of  Economics  and  Statistics,  Government  of  Kerala,  and  are 
considered as the main dataset for this study. 	
 	
4.1	 	Data Partitioning	 	 
The Entire dataset is partitioned into 2 parts: for example, say, 75% of the dataset is 	
used for training the model and 25% of the data is set aside to test the model.   
5.	 	Tools and Libraries Used 
 
For  the  current  project,  Python  is  chosen  as  the  programming  language  for  all  the 	
implementations,  starting  from  extracting  the  data,  to  evaluating  the  model.  It  has  a  huge 
library support for applications in the field of Machine Learning and Artificial Intelligence 
and this makes Python more suitable for solving problems in real world scenarios. Google 
Collab  notebook  were  used  to  write  the  code  for  the  project.  All  the  exploratory  data 
analysis  was  done  using  Python  libraries  like  NumPy,  Pandas,  Matplotlib,  and  Seaborn. 
The  selection,  training  and  evaluating  the  model  was  done  using  the  Scikit-Learn  library 
and its classes. No specific operating system is required as Python is a portable language. 	
 	
5.1	 	Performance Parameters  	
    In  this  study,  the  performance  parameters  used  are  RMSE  and  R2  score.  As  a 
supervised  learning  algorithm  is  being  used,  the  training  instances  have  predefined  labels. 
The RMSE is a typical performance measure for regression tasks, providing an estimate of 
the  system's  prediction  error,  with  larger  errors  being  given  a  higher  weight.  A  higher 
RMSE  value  indicates  lower  efficiency  of  the  model.  Another  evaluation  metric  used  in 
this study is the R2 score, which is a popular measure of the model's accuracy. It represents 
how  closely  the  data  values  align  with  the  regression  line.  A  higher  value  of  R2  or  (R) 
indicates a better fit of the model to the data.   
6.	 	Results and Analysis	 	
 
The  following  section  outlines  the  findings  of  our  study,  which  involved 	
implementing  the  model  discussed.  Prior  to  implementation,  we  conducted  a  thorough 
analysis of the dataset to gain insights. Therefore, this section firstly provides the statistical 
results and followed by the performance analysis of the proposed model.

5 	
Compendium  of  Project  Reports 
6.1	 	Statistical Analysis 	
 	
Analyzing the Area of Paddy Cultivation by District and Season between   	
  2015-2020

6                  	Compendium  of  Project  Reports	 	
 	
 
 	
 	   Paddy  cultivation  in  most  districts  typically  occurs  during  the  winter  season,  but 	
in  Alappuzha  and  kottayam,  it  takes  place  during  the  summer  season,  and  there 
has been a progressive increase in cultivation. 	
 	   In  kerala  Palakkad  has  the  largest  paddy  area  over  15  years  but  the  same  time 	
there  has  been  a  gradual  reduction  in  overall  paddy  cultivation  area  during 
Autumn and winter. 	
 	
Analyzing the Area of Tapioca Cultivation by District and Season between  	
2015-2020

7 	
Compendium  of  Project  Reports 	
 	
 
 	
 	
 	
 	
Tapioca  cultivation occurs mostly during the summer season. Thiruvananthapuram 	
and  Kollam  are  the  leading  districts  for  tapioca  cultivation.  In  Alappuzha,  Idukki,  and

8                  	Compendium  of  Project  Reports	 	
 
Palakkad,  there  has  been  a  slight  increase  in  the  area  of  tapioca  cultivation  during  the 
summer season. 
 
6.2	 	Statistical Description of the Dataset 
 	
Paddy 	
 	
index  Area  Production  Rainfall 
std	 	22456.5  59881.1  749.8 	
min	 	603.0  1467.0  43.7 	
mean	 	15234.8  39940.3  2873.7 	
max	 	113919.0  270103.0  5895.0 	
count	 	210.0  210.0  210.0 	
75%	 	17275.25  49053.0  3351.60 	
50%	 	6184.5  13472.5  2903.75 	
25%	 	2873.5  6392.5  2379.125 	
 	
Tapioca 	
 	
index  Area  Production  Rainfall 
count	 	224.0  224.0  224.0 	
mean	 	5266.6  185509.56  2903.9 	
std	 	4979.66  160725.89  753.72 	
min	 	245.0  5672.0  43.7 	
25%	 	1815.75  61252.5  2385.42 	
50%	 	4088.5  134460.0  2940.45 	
75%	 	6128.75  241929.25  3413.82 	
max	 	23814.0  694984.0  5895.0 	
 
 	
6.3	 	Correlation matrix for the training Set  	
Rice 	
 	
index  Area  Production	 	Rainfall 	
Rainfall	 	-0.17  -0.18  1.0 	
Production	 	0.99  1.0  -0.18 	
Area	 	1.0  0.99  -0.17 	
  	
Tapioca 	
 	
 	index  Area  Production  Rainfall 	
Area	 	1.0  0.96  -0.31 	
Production	 	0.96  1.0  -0.27 	
Rainfall	 	-0.31  -0.27  1.0

9 	
Compendium  of  Project  Reports 
6.4	 	Heat map showing the correlation between numerical attributes training    
set 	
 
 
6.5	 	Scatter Matrix of the Training Set 	
 	
Paddy 	
 	
 	
 	
Tapioca 	
 	
Rice  Tapioca

10                  	Compendium  of  Project  Reports	 	
 	
 	 	
From all the above tables and diagrams correlation between the area, production of 	
paddy, tapioca and rainfall indicates a weak negative correlation between the two variables. 
The  relationship  is  not  very  strong.  Also  its    important  to  note  that  correlation  does  not 
imply causation, and there may be other factors that affect the area of paddy, tapioca, such 
as soil type, temperature, or human factors like land use changes or agricultural practices. 
  
6.6	 	Evaluation of Scores 
 
In  this  paper,  the  algorithms  such  as  Linear  Regression  and  Support  Vector 	
Regressor are used to predict rice and Tapioca yield. The mean square error has been used 
to measure the loss incurred by the model. 	
Rice 	
 	
Model  RMSE         R2 	
Linear Regression	 	10220  0.98 	
Support Vector Regression	 	52742  0.28 	
      	 	
Tapioca 	
 	
Model  RMSE         R2 	
Linear Regression	 	40570  0.93 	
Support Vector Regression	 	38747  0.04 	
 
 	This  result  clearly  shows  that	 Linear  Regression  fared  much  better  than  Support 	
vector regression.

11 	
Compendium  of  Project  Reports 
 
Conclusion	 	
 
The  Linear  Regression  is  the  basic  regression  algorithm  which  can  be  used  only 	
when the dataset is simple.  It has been observed that that performance of linear regression 
has  been  affected  by  non-linearity  of data,  presence  of  outlier  and  high  correlation  among 
the features. Therefore, it may not work well for complex datasets. For this algorithms like 
Ridge  Regression  and  Lasso  Regression,  which  are  more  regularized  forms  of  Linear 
Regression  can  be  used  when  the  dataset  contains  a  large  number  of  features  or  unwanted 
features.  Predicting  agricultural  production  involves  analyzing  various  factors  such  as, 
weather  patterns,  soil  quality,  and  agricultural  practices.  Statistical  models  can  be  used  to 
incorporate data from multiple sources, and machine learning algorithms  can be trained to 
make  predictions  about  future  production.  However,  it's  important  to  note  that  predicting 
agricultural  production  is  subject  to  uncertainties  such  as  unforeseen  weather  events,  pest 
outbreaks,  or  market  fluctuations,  and  predictions  should  be  regularly  monitored  and 
updated throughout the growing season. As mentioned in the introduction, if we have data 
on the consumption of rice and tapioca by Keralite, we can inform policymakers about the 
current  status  of  paddy  and  tapioca  farming  in  Kerala.  This  information  can  help  them 
develop effective policies and plans to improve the agricultural sector, increase crop yields, 
and support farmers. 	
 As  the  future  aspects,  we  will  work  to  evaluate  non-linear  techniques  and  more 	
advanced  deep  learning  methods  can  be  used  as  they  have  the  advantage  during  feature 
selection phase in machine learning.	 	
 
 
 	
References 	
 	
 
 	 	
   
 
 
 	
[1]  A.P.S  Manideep  and  Dr.  Seema  Kharb,â  A  Comparative  Analysis    of  Machine  Learning 	
Prediction  Techniques  for  Crop  Yield  Prediction  in  Indiaâ,  Turkish  Journal  of  Computer  and 
Mathematics Education Vol.13 No.02 (2022), 120-133 	
[2]  Dr. B M Sagar, Dr.N K Cauvery, Dr.Padmashree T. and Dr. R.     	
Rajkumar,  âRice  and  Wheat  yield  prediction  in  india  using  decision  tree  and  Random  forestâ, 
Computational intelligence and Machine Learning, Vol3, issue2, 2022,PP 1-8. 	
[3]  Mayank  champaneri,  Darpan  Chachpara,  Chaitanya      Chandvidkar  and  mansing  Rathod, 	
âCrop  Yield  Prediction  Using  Machine  Learningâ,  Interational  Journal  on  Science  and 
Research, ISSN:2319-7064, 2019

12                  	Compendium  of  Project  Reports

13 	
Compendium  of  Project  Reports 	
Prediction of Cost of cultivation of 	
important crops in Kerala using Machine 	
Learning	
 	
Submitted  By 	
Smt. Shailamma K., 	
Deputy Director

14                  	Compendium  of  Project  Reports	 	
 
Introduction 
 	
India  is  predominantly  an  agricultural  nation,  with  60%  of  the  population 	
depending on agriculture for a living in one way or another. In emerging nations like India, 
where  agriculture  serves  as  the  primary  driver  of  economic  growth,  the  development  of 
agriculture  is  crucial.    The  development  of  agriculture  is  made  feasible  by  increased 
investment  to  satisfy  the  rising  capital  requirements  of  contemporary  agriculture.  About   
58%  of  India's  population  relies  primarily  on  agriculture  as  a  source  of  income,  making 
India  one  of  the  key  players  in  the  global  agricultural  industry.  Families  are  typically 
reliant  on  agriculture.  Agriculture  accounts  for  a  significant  portion  of  the  nation's  GDP. 
To  achieve  the  strict  standards,  agricultural  practices  must  be  modernised.  Prices  for  the 
crop have fluctuated significantly over the last few years. As a result the frequency of crop 
damage  has  increased.  This  prediction  system's  primary  goal  is  to  give  farmers  a  better 
understanding of the cost of cultivation and how to maximise produce. 	
 	
Objective 
 	
The  most  significant  and  prominent  issue affecting  Indian  agriculture  at  the  moment 	
is  the  rise  in  cultivation  costs.  The  cost  of  cultivation  is  made  up  of  a  variety  of  factors, 
including  personnel  costs,  equipment  rental  fees,  costs  for  seeds,  fertilizer,  and  pesticides, 
as well as irrigation costs. Profitability and the cost-return ratio are directly impacted. Cost 
of  cultivation  of  major  crops  are  very  essential  for  formulating  proper  support  price 
policies,  creating  marketing  facilities  and  assessing  loss  out  of  natural  calamities,  and  the 
share of agriculture in GDP. We need reliable data on crop husbandry right from sowing to 
harvesting  stages.  For  this,  the  department  conducts  an  annual  survey  on  Cost  of 
Cultivation of Important Crops in Kerala during every agricultural  year (July to June). The 
major  components  of  the  costs  involved  are  seed/seedlings,  fertilisers,  labour,  rent, 
equipment,  irrigation  charges  etc.	 The  main  objective  of  the  survey  is  to  produce  reliable 	
estimates  on  production  cost  involved  in  major  agricultural  crops  in  the  State.  Major 
perennial,  annual  and  seasonal  crops  are  covered  in  the  survey  with  a  breakup  of  cost 
incurring during various stages from sowing to harvest. 	
 	
 Methodology 
 	
The  survey  covered  all  the  districts  in  the  state  by  considering  taluk  as  a  stratum.  	
From  each  Taluk,  required  numbers  of  investigator  zones  were  selected  using  circular 
systematic sampling method. From selected zones, cultivators and  corresponding holdings 
are  selected.  The  holdings  are  grouped  under  three  size  classes  viz;  small,  medium  and 
large  according  to  the  area.  The  investigators  visited  the  selected  holdings/cultivators  and 
collected the required information in the prescribed format. 	
 
Cost incurred for growing the selected crops are classified under coat âAâ, cost âBâ, 	
and  cost  âCâ.  Analysis  of  the  data  is  carried  out  based  on  cost  âAâ.  Cost  âAâ  is  estimated 
based  on  all  kind  of expenses(paid  out  cost)  actually  incurred  by  the cultivators  includes  , 
Hired  human  labour,  Animal  labour,  Machine  labour,  Seed/  seedlings,  Farm  yard  Manure 
and  Chemical  fertilizers,  Plant  protection,  Land  tax  and  Irrigation  Cess,  Repair  and 
maintenance  charges  of  implements,  machinery  and  buildings,  Interest  on  working  capital 
and Other expenses.

15 	
Compendium  of  Project  Reports 	
 	
Proposed Methodology 
 
Many factors are influential in agriculture, especially in cultivation and production. In this 
work,  the  impact  of  different  factors  that  affect  the  cost  of  cultivation  is  considered  and 
predicts  how  these  factors  affect  the  cost  and  forecast  the  cost.  For  that  we  use  linear 
regression based prediction models. It is a machine learning algorithm based on supervised 
learning.  It  is  used  for  finding  out  the  relationship  between  variables  and  forecasting. 
Linear  regression  is  a  powerful  tool  for  understanding  and  predicting  the  behavior  of  a 
variable.  It  is  a  statistical  approach  for  modeling  the  relationship  between  a  dependent 
variable and a given set of independent variables. 
  
Method Used 
 	
The  cost  of  cultivation  is  influenced  by  more  than  two  items.  So  we  use  multiple  	
linear  regression  based  prediction  models  .Multiple  linear  regression  (MLR),  also  known 
simply  as  multiple  regression,  is  a  statistical  technique  that  uses  several  independent 
variables to predict the outcome of a dependent variable. 
 
  	yi=Î²0+Î²1xi1+Î²2xi2+...+Î²pxip+Ïµ   for i=1,2,â¦,n observations 	
    where,     yi=dependent variable 
            xi=independent variables 
            Î²0=y-intercept (constant term) 
            Î²p=slope coefficients for each independent variable 
            Ïµ=the modelâs error term (also known as the residuals) 
 
Dataset Used 
 	
The  actual  cost  incurred  by  the  cultivators  during  2019-20  is  considered  for 	
the study. 	
Cost incurred for human labour, seed, fertilizer and other expenses are taken 	
into consideration. 	
 
Farmer_id  Cost_fertilizer	 	Cost_seed  Cost_human 	Cost_other 
expenses 	Cost A 	
22797	 	16487.25  2964  30010.5  23465  72926.75 	
22799	 	10497.5  2315.63  31122  16981.25  60916.38 	
22800	 	5631.6  2667.6  5928  13585  27812.2 	
22801	 	12473.5  2593.5  39421.2  8645  63133.2 	
22802	 	11559.6  2964  33888.4  10374  58786 	
22803	 	11732.5  2717  35568  13173.33  63190.83 	
22805	 	11423.75  2778.75  51005.5  13585  78793 	
22806	 	12350  3334.5  13832  12350  41866.5 	
22807	 	11032.67  2717  39190.67  26676  79616.33 	
22810	 	18278  2964  82596.8  11362  115200.8 	
22811	 	21538.4  2964  41100.8  12844  78447.2

16                  	Compendium  of  Project  Reports	 	
 	
22969	 	258.83  1331.14  42049.16  0  43639.13 	
22970	 	2074.8  3334.5  24700  0  30109.3 	
22971	 	5086.37  3805.63  35037.41  2287.04  46216.44 	
22972	 	7782.75  2020.91  64556.82  8532.73  82893.2 	
22973	 	6249.1  2223  57427.5  10703.33  76602.93 	
22974	 	5705.7  2470  35074  3737.93  46987.63 	
22975	 	6570.2  2470  40960.83  6998.33  56999.37 	
22976	 	6995.04  2964  47177  4446  61582.04 	
22977	 	6007.04  2964  67431  988  77390.04 	
22978	 	6082.38  3087.5  10291.67  2058.33  21519.87 	
23004	 	6340.12  2774.94  16924.07  2363.27  28402.41 	
23005	 	5132.97  2881.67  10806.25  2315.63  21136.51 	
23006	 	6101.93  2593.5  14326  14614.17  37635.6 	
23007	 	7696.79  2761.6  39468.54  1440.83  51367.77 	
23008	 	9221.33  2815.8  34621.17  3087.5  49745.8 	
23009	 	6873.96  2851.01  21428.56  1182.45  32335.98 	
23010	 	5966.33  2861.79  28669.03  553.62  38050.78 	
23011	 	6923.72  2809.63  26019.91  14279.69  50032.94 	
23012	 	4802.78  2641.53  32212.92  1543.75  41200.97 	
23013	 	7512.92  2737.58  15725.67  926.25  26902.42 	
23014	 	7913.88  2849.76  11533.67  11713.98  34011.28 	
23015	 	4433.46  2836.64  26147.27  2026.17  35443.54 	
23016	 	10229.92  2881.67  27664  4528.33  45303.92 	
23017	 	7426.84  2475.61  28685.68  1571.82  40159.95 	
23018	 	8026.06  2613.6  37006.92  2584.88  50231.47 	
23063	 	17819.29  3920.63  2352.38  11467.86  35560.16 	
23065	 	0  4116.67  9495.78  8789.08  22401.53 	
23066	 	9297.64  4016.26  1606.5  22015.8  36936.21 	
23068	 	11866.19  4074.23  25973.2  12318.17  54231.78 	
23069	 	9756.5  3952  12967.5  24700  51376 	
23070	 	13959.03  3952  8962.57  11143.23  38016.83 	
23071	 	7431.48  4295.65  0  10943.17  22670.3 	
23072	 	11967.15  3892.12  8350.1  18510.03  42719.4 	
23073	 	2720.09  3952  29281.85  9555.81  45509.75 	
23074	 	11942.45  3952  11244.68  27089.73  54228.85 	
23075	 	15510.54  4249.46  15669.89  10437.74  45867.63 	
23077	 	23481.47  4610.67  0  11444.33  39536.47 	
23079	 	11761.32  3952  22271.17  25373.08  63357.56 	
23080	 	13667.33  4116.67  14964.08  10960.63  43708.71 	
23081	 	2346.5  3425.07  36007.11  5488.89  47267.57 	
23082	 	1743.11  2935.77  35709.14  8821.43  49209.46 	
23083	 	0  0  46899.13  9262.5  56161.63 	
23084	 	18125.17  2964  77774.13  4631.25  103494.5 	
23085	 	11826.36  3003.52  0  3705  18534.88 	
23086	 	4446  3249.42  34429.06  5488.89  47613.37 	
23087	 	18792.16  2903.51  29161.12  2520.41  53377.2

17 	
Compendium  of  Project  Reports 	
23088	 	1437.54  2845.44  40186.9  12350  56819.88 	
23089	 	1466.56  3260.4  35552.56  3087.5  43367.03 	
23090	 	12331.48  3062.8  40369.06  3087.5  58850.84 	
23091	 	19698.25  2964  48473.75  9262.5  80398.5 	
23092	 	16697.2  2529.28  36284.3  4940  60450.78 	
23093	 	1501.76  3003.52  36160.8  4940  45606.08 	
23094	 	16385.98  3319.68  25411.36  3952  49069.02 	
23095	 	1501.76  3003.52  39816.4  4940  49261.68 	
23188	 	6076.2  1704.3  19216.6  449.54  27446.64 	
23189	 	7558.2  1704.3  19043.7  5685.94  33992.14 	
23190	 	7551.14  1693.71  18313.29  6588.73  34146.87 	
23191	 	3705  2321.8  24041.33  2914.6  32982.73 	
23192	 	7753.06  2305.33  31698.33  5149.26  46905.99 	
23193	 	3705  2778.75  24082.5  2192.13  32758.38 	
23194	 	6058.76  1721.74  19760  6871.69  34412.19 	
23195	 	6175  2297.1  24453  6619.6  39544.7 	
23196	 	3705  2702.47  26734.12  12684.18  45825.76 	
23197	 	6422  2334.15  29393  6014.45  44163.6 	
23198	 	9962.33  2321.8  18936.67  444.6  31665.4 	
23199	 	6833.67  2766.4  30298.67  345.8  40244.53 	
23200	 	7934.88  2778.75  24082.5  339.63  35135.75 	
23201	 	12028.9  1704.3  19019  2820.74  35572.94 	
23202	 	5681  2717  32110  2799.33  43307.33 	
23280	 	5448.97  5987.88  65866.67  16092.42  93395.94 	
23281	 	6422  1915.51  58032.4  19029.08  85398.99 	
23282	 	2825.68  3952  54710.5  17784  79272.18 	
23283	 	5779.8  3952  51623  17784  79138.8 	
23284	 	12844  4149.6  54710.5  19918.08  91622.18 	
23285	 	6296.08  7749.02  53637.75  17435.29  85118.14 	
23286	 	8562.67  7978.1  37667.5  8892  63100.27 	
23287	 	7706.4  9386  40137.5  8892  66121.9 	
23288	 	2989.44  2815.8  40137.5  15067  61009.74 	
23289	 	7339.43  2681.71  35638.57  9668.29  55328 	
23290	 	9454.61  4665.56  37211.24  8892  60223.4 	
23291	 	2607.92  3009.14  12224.62  13541.12  31382.79 	
23292	 	6422  4347.2  37358.75  6158.53  54286.48 	
23293	 	4587.14  5645.71  35973.79  7621.71  53828.36 	
23294	 	6534.67  3293.33  36725  9360  55913 	
23332	 	91.08  1311.5  39651.15  0  41053.73 	
23623	 	7372.95  6402.24  18649.41  1469.65  33894.25 	
23624	 	7459.4  7706.4  31617.98  2692.3  49476.08 	
23625	 	9784.91  5157.36  32296.63  1512.88  48751.77 	
23626	 	10277.94  8892  0  1578.06  20748 	
23627	 	7788.38  5549.62  27959.54  6148.72  47446.26 	
23628	 	7968.22  6520.8  19143.44  1296.75  34929.21 	
23629	 	11271.43  5730.4  23644.4  1337.92  41984.15

18                  	Compendium  of  Project  Reports	 	
 	
23630	 	7254  7488  30854.88  2210  47806.88 	
23791	 	6103.15  5119.64  15045.2  1212.55  27480.52 	
23792	 	11411.4  0  16755.57  7698.17  35865.13 	
23825	 	9781.2  3293.33  53887.17  13132.17  80093.87 	
23828	 	8774.04  3226.12  28127.76  13443.86  53571.78 	
23838	 	9298.82  2847.76  74535.88  15037.94  101720.4 	
23841	 	5763.33  3046.33  66072.5  3972.58  78854.75 	
23854	 	4952.35  2881.67  52075.83  2202.42  62112.27 	
23855	 	11223.97  3359.93  75552.94  2324.71  92461.54 	
23857	 	9725.63  2701.56  61557.03  2470  76454.22 	
23859	 	2629.35  2948.06  31538.98  13651.4  50767.8 	
23862	 	3112.2  2855.32  53144.52  14128.4  73240.44 	
23864	 	4418.56  2881.67  65500.74  1738.15  74539.11 	
23866	 	5804.5  2881.67  0  2264.17  10950.33 	
23867	 	10003.5  3046.33  64014.17  3355.08  80419.08 	
23869	 	12903.62  2981.03  66093.79  2768.1  84746.55 	
23871	 	12926.33  2788.36  41715.56  1976  59406.24 	
23872	 	6327  2992.5  28500  1805  39624.5 	
23899	 	6125.6  2173.6  17425.85  14820  40545.05 	
23900	 	6379.04  2244.48  33973.24  23529.43  66126.2 	
23901	 	6520.8  2309.45  17425.85  15301.65  41557.75 	
23902	 	6298.5  2547.19  29832.2  17158.78  55836.67 	
23903	 	6520.8  2445.3  29769.68  24712.35  63448.13 	
23905	 	6549.91  2425.89  30023.73  16187.32  55186.86 	
23908	 	6276.71  2397.35  27818.01  11362  47854.07 	
23909	 	6257.33  2445.3  19899.97  20826.22  49428.82 	
23910	 	6520.8  2445.3  27393.54  17604.93  53964.56 	
23911	 	6422  2377.38  30498.33  25777.54  65075.24 	
23912	 	6520.8  2445.3  49350.6  10262.85  68579.55 	
23913	 	6586.67  2508  34341.09  18373  61808.75 	
24005	 	7765.8  3445.65  30779.7  1136.2  43127.35 	
24006	 	6648.42  3556.8  0  105427.8  115633.1 	
24007	 	8082.72  3968.07  30372.97  18449.7  60873.45 	
24008	 	7904  3445.65  27980.47  24230.7  63560.82 	
24012	 	8068.67  3747.08  26358.1  17518.7  55692.55 	
24013	 	8001.5  5190.9  28762.5  23868  65822.9 	
24015	 	7931.68  3570.43  41628.02  9633  62763.13 	
24016	 	8016.27  3563.99  21837.05  18188.18  51605.49 	
24021	 	7632.3  4075.5  23641.43  10091.71  45440.94 	
24024	 	8080.43  3952  39685.4  9728.27  61446.1 	
24103	 	4890.6  1852.5  24482.64  13338  44563.74 	
24104	 	6175  1789.86  34723.19  4474.64  47162.68 	
24105	 	3803.8  1914.25  26379.6  13338  45435.65 	
24106	 	4693  1837.68  25564.5  13585  45680.18 	
24122	 	3774.36  3981.94  22142.68  16573.95  46472.93 	
24123	 	2543.9  3919.61  30551.07  28166.1  65180.67

19 	
Compendium  of  Project  Reports 	
24124	 	2964  3952  13554.13  22971  43441.13 	
24125	 	2136.55  4149.6  41002  18599.1  65887.25 	
24126	 	2408.25  3952  10464.57  22806.33  39631.15 	
24127	 	1905.43  4327.44  30204.57  16549  52986.44 	
24128	 	2124.2  4149.6  43348.5  18426.2  68048.5 	
24129	 	2321.8  11065.6  27417  19019  59823.4 	
24130	 	1994.53  4149.6  31893.88  16437.85  54475.85 	
24131	 	3735.88  3952  17266.84  26398.13  51352.84 	
24132	 	1778.4  4149.6  59156.5  19537.7  84622.2 	
24133	 	1778.4  4034.33  11449.48  15787.99  33050.2 	
24134	 	1970.51  3872.96  22559.33  16933.22  45336.03 	
24135	 	2390.96  3651.65  47374.6  15264.6  68681.81 	
24136	 	3237.46  4149.6  35409.21  24700  67496.28 	
24291	 	8953.75  2223  27664  0  38840.75 	
24292	 	8424.46  2434.71  13408.57  4410.71  28678.46 	
24293	 	8205.89  2209.28  18223.11  0  28638.28 	
24294	 	9185.31  2223  12967.5  0  24375.81 	
24295	 	9553.09  2615.29  9298.82  4358.82  25826.03 	
24296	 	7958.89  2209.28  17454.67  0  27622.83 	
24297	 	8584.76  2216.98  21808.29  4819.51  37429.54 	
24298	 	7862.83  2470  22888.67  0  33221.5 	
24387	 	1818.4  1591.1  35958.96  20078.22  59446.69 	
24388	 	8595.6  1630.2  25490.4  33345  69061.2 	
24389	 	3705  1552.57  31492.5  7851.07  44601.14 	
24408	 	1852.5  1698.13  20377.5  1929.69  25857.81 	
24411	 	30737.78  1536.89  114388.4  2195.56  148858.7 	
24419	 	0  3175.71  0  793.93  3969.64 	
24423	 	14166.18  3196.47  19372.55  1816.18  38551.37 	
24425	 	12350  1605.5  42237  1852.5  58045 	
24426	 	7591.62  4358.82  37776.47  1343.97  51070.88 	
24507	 	10574.69  2223  21921.25  0  34718.94 	
24508	 	13210.34  2331.24  15819.1  0  31360.67 	
24792	 	6916  2815.8  26182  419.9  36333.7 	
24793	 	7853.33  1425  48133.33  285  57696.67 	
24794	 	9914.79  2435.21  25047.89  226.13  37624.01 	
24795	 	12411.75  2686.13  22230  308.75  37636.63 	
24796	 	7706.4  2213.12  41496  395.2  51810.72 	
24797	 	16631.33  1976  23876.67  205.83  42689.83 	
24798	 	10081.63  2117.14  30244.9  504.08  42947.76 	
24799	 	4322.5  2778.75  28713.75  370.5  36185.5 	
24800	 	7904  2305.33  16466.67  0  26676 	
24801	 	10492.56  2578.68  29047.2  395.2  42513.64 	
24802	 	8315.67  2223  16796  230.53  27565.2 	
24803	 	9445.71  2328.86  59985.71  271.43  72031.71 	
24943	 	2408.25  2686.13  28405  0  33499.38 	
24945	 	7059.26  2667.6  14573  247  24546.86

20                  	Compendium  of  Project  Reports	 	
 	
24966	 	6027.4  5142.36  3809.16  0  14978.91 	
24967	 	6452.88  5335.2  13832  6175  31795.08 	
24968	 	7821.67  5928  0  0  13749.67 	
24969	 	6211.32  6538.24  17798.53  11805.15  42353.24 	
24970	 	7671.25  5335.2  5890  9025  27921.45 	
24971	 	6100.9  5409.3  9880  9056.67  30446.87 	
24972	 	6257.33  5335.2  17652.27  12350  41594.8 	
24973	 	6399.55  5254.36  21197.09  12350  45201 	
24974	 	6768.75  3420  4560  11875  26623.75 	
24975	 	6704.29  12702.86  0  16937.14  36344.29 	
24976	 	6114.12  5009.58  21360.28  7827.46  40311.44 	
24977	 	5991.06  3783.83  0  10510.64  20285.53 	
24978	 	6526.2  4168.13  53124.3  13507.81  77326.44 	
24979	 	6801.45  5154.78  9593.62  9665.22  31215.07 	
24980	 	6650.51  5706.1  12997.22  9906.42  35260.24 	
25000	 	9255.09  2037.75  17808.7  13313.3  42414.84 	
25060	 	9818.25  2470  54447.39  2470  69205.64 	
25061	 	9818.25  2470  65208  2470  79966.25 	
25062	 	9818.25  2470  84844.5  2470  99602.75 	
25063	 	9818.25  2470  0  16584.29  28872.54 	
25064	 	9818.25  2470  68748.33  14820  95856.58 	
25065	 	9818.25  2470  133544.7  2470  148302.9 	
25066	 	9818.25  2470  56995.25  2470  71753.5 	
25067	 	9818.25  2470  50006.67  2470  64764.92 	
25068	 	9818.25  2470  65193.03  2470  79951.28 	
25069	 	9818.25  2470  5513.04  11757.2  29558.49 	
25070	 	9818.25  2470  100446.7  11801.11  124536 	
25071	 	9818.25  2470  72524.31  11413.1  96225.66 	
25072	 	9818.25  2470  61926.43  11644.29  85858.96 	
25073	 	9818.25  2470  41166.67  11879.52  65334.44 	
25074	 	9818.25  2470  41707.71  12350  66345.96 	
25164	 	9061.61  2766.4  24288.33  3869.67  39986.01 	
25165	 	7449.52  2558.92  18870.8  17413.5  46292.74 	
25166	 	13399.75  3087.5  30566.25  5094.38  52147.88 	
25167	 	23897.25  2766.4  18772  14869.4  60305.05 	
25168	 	4949.88  2766.4  10472.8  12646.4  30835.48 	
25169	 	12010.38  3087.5  34425.63  12890.31  62413.81 	
25170	 	6669  0  21406.67  10497.5  38573.17 	
25171	 	6669  2593.5  2305.33  11279.67  22847.5 	
25172	 	9600.07  2334.15  19142.5  15972.67  47049.38 	
25173	 	7281.56  2593.5  26774.8  1976  38625.86 	
25174	 	7133.36  2845.44  21637.2  12295.66  43911.66 	
25175	 	8336.25  2334.15  25243.4  8348.6  44262.4 	
25176	 	9880  2694.55  17514.55  14056.55  44145.64 	
25177	 	12610.72  3293.33  22778.89  12830.28  51513.22 	
25257	 	4663.36  4742.4  17981.6  38927.2  66314.56

21 	
Compendium  of  Project  Reports 	
25258	 	1963.65  5928  38532  12350  58773.65 	
25259	 	3290.04  6916  39520  30134  79860.04 	
25260	 	7904  4446  70642  20748  103740 	
25261	 	4683.12  7904  49400  33098  95085.12 	
25262	 	6051.5  5928  42484  26676  81139.5 	
25263	 	8892  7904  64714  30134  111644 	
25264	 	12350  3952  55328  44830.5  116460.5 	
25265	 	9484.8  4763.57  33874.29  7057.14  55179.8 	
25266	 	17191.2  4446  83980  37050  142667.2 	
25267	 	16466.67  4940  75746.67  10374  107527.3 	
25418	 	4831.03  3632.35  45767.65  27605.88  81836.91 	
25419	 	4364.5  3733  7839.29  18988.51  34925.3 	
25420	 	4285.45  4116.67  46106.67  27170  81678.78 	
25421	 	4237.17  3761.14  3929.55  29009.03  40936.88 	
25422	 	4701.23  3705  7492.33  27894.53  43793.1 	
25423	 	3798.86  3754.4  5532.8  28098.72  41184.78 	
25424	 	4363.67  3705  6916  28322.67  43307.33 	
25425	 	4024.2  3762  4788  28473.4  41047.6 	
25426	 	4010.22  3757.93  4940  28574.37  41282.52 	
25427	 	5051.15  3705  8645  28429.7  45830.85 	
25428	 	5219.93  3705  10374  27680.47  46979.4 	
25429	 	4664.18  3705  7492.33  19554.17  35415.68 	
25430	 	4557.15  3705  6916  28108.6  43286.75 	
25431	 	3835.91  3754.4  5532.8  27970.28  41093.39 	
25432	 	5878.6  3705  8068.67  28322.67  45974.93 	
25535	 	8637.59  2000.7  31214.63  9311.9  51164.82 	
25539	 	7150.65  1896.96  22674.6  49.4  31771.61 	
25541	 	9434.38  1909.79  42626.6  12477.32  66448.09 	
25542	 	0  2533.33  48070  0  50603.33 	
25544	 	9305.73  2521.46  22044.75  51.46  33923.39 	
25545	 	6800.52  1847.69  37916.1  12401.32  58965.64 	
25547	 	5063.5  2315.63  35104.88  185.25  42669.25 	
25548	 	10739.93  2515.74  37461.67  274.44  50991.78 	
25549	 	16631.33  2352.38  18230.95  235.24  37449.9 	
25550	 	7976.04  2408.25  37976.25  0  48360.54 	
25551	 	9262.5  2408.25  45942  164.67  57777.42 	
25552	 	6986.57  2470  32833.36  617.5  42907.43 	
25554	 	11115  2247.7  30875  247  44484.7 	
25562	 	2470  2470  20377.5  1852.5  27170 	
25566	 	22293.33  2964  22800  506.67  48564 	
25656	 	5335.2  5928  87932  8398  107593.2 	
25657	 	12844  3952  74100  21242  112138 	
25658	 	5310.5  3087.5  84597.5  6175  99170.5 	
25659	 	10406.93  0  50223.33  14243.67  74873.93 	
25798	 	12012  3952  47125  11050  74139 	
25799	 	8172  3952  35444.5  31282.55  78851.05

22                  	Compendium  of  Project  Reports	 	
 	
25800	 	8343.97  3952  54610.16  31801.25  98707.38 	
25801	 	8966.24  3952  0  31353.29  44271.54 	
25802	 	12254.45  3923.77  46102.55  23994.29  86275.05 	
25803	 	9305.73  3952  35444.5  34086  82788.23 	
25804	 	10324.35  3952  35451  32500  82227.35 	
25805	 	11844.93  3952  35285.71  33874.29  84956.93 	
25806	 	10151.44  3952  35435.49  32933.33  82472.27 	
25807	 	9071.32  3952  35444.5  21983  70450.82 	
25808	 	12526.43  3952  34844.64  25546.86  76869.93 	
25809	 	11584.3  3952  35321  36062  86919.3 	
25810	 	13128.67  3952  35043.13  21365.5  73489.29 	
25811	 	11998.19  3952  34965.94  19358.63  70274.75 	
25812	 	8916.7  3952  35444.5  34333  82646.2 	
26105	 	9163.7  4940  20501  5975.75  40580.45 	
26106	 	8284  5700  28215  6692.75  48891.75 	
26107	 	8645  4940  39211.25  7799.03  60595.28 	
26108	 	8645  4940  33437.63  7332.81  54355.44 	
26109	 	7330.96  4940  29244.8  2302.04  43817.8 	
26110	 	9447.11  5092.78  25379.04  19745.57  59664.51 	
26111	 	9163.7  4940  36556  8195.46  58855.16 	
26112	 	9027.05  5330.94  26210.43  6022.18  46590.6 	
26113	 	5873.11  5488.89  29228.33  6729.93  47320.26 	
26114	 	7780.5  4940  30504.5  6589.96  49814.96 	
26115	 	7330.96  4940  26676  26219.54  65166.5 	
26116	 	9163.7  4940  24823.5  7528.56  46455.76 	
26117	 	7868.71  5645.71  29675.29  7062.79  50252.5 	
26118	 	9163.7  5269.33  21900.67  8564.31  44898.01 	
26119	 	6125.6  4940  31430.75  26542.62  69038.97 	
26175	 	17868.91  0  42067.19  4564.35  64500.45 	
26176	 	10806.25  4322.5  41990  7286.5  64405.25 	
26177	 	7179.47  4610.67  25655.07  3392.13  40837.33 	
26178	 	14731.79  0  40508  16363.75  71603.54 	
26179	 	14608.29  0  71136  5469.29  91213.57 	
26180	 	14981.73  3675.6  32462.86  15158.15  66278.33 	
26181	 	14126.15  2694.55  26747.85  21329.57  64898.13 	
26182	 	18209.78  0  35283.36  21503.11  74996.26 	
26183	 	20289.29  0  34756.43  7210.05  62255.76 	
26184	 	11171.46  0  39971.66  13479.14  64622.26 	
26185	 	0  0  22543.27  0  22543.27 	
26186	 	14491.64  3051.18  31479.42  21358.24  70380.47 	
26187	 	15869.75  3952  26589.55  21242  67653.3 	
26537	 	3378.96  3458  23341.5  3952  34130.46 	
26539	 	3487.64  3458  17784  3458  28187.64 	
26544	 	3375.26  3458  13338  3458  23629.26 	
26545	 	3202.44  3458  18854.33  3952  29466.77 	
26546	 	2818.89  3396.25  21689.69  3859.38  31764.2

23 	
Compendium  of  Project  Reports 	
26547	 	3406.62  3458  24020.75  2964  33849.37 	
26549	 	3448.12  3458  22724  2881.67  32511.79 	
26550	 	3604.22  3458  26305.5  4446  37813.72 	
26552	 	3380.81  3458  21118.5  0  27957.31 	
26553	 	3371.55  3458  25564.5  3952  36346.05 	
26555	 	3029.87  3458  25276.33  2634.67  34398.87 	
26557	 	3195.56  3458  24206  2284.75  33144.31 	
26559	 	2252.64  3359.2  20105.8  2371.2  28088.84 	
26560	 	2704.65  3458  19760  3458  29380.65 	
26594	 	2872.26  3528.57  21594.86  2540.57  30536.26 	
26603	 	13008.67  2881.67  52899.17  4446  73235.5 	
26608	 	16820.8  2532.01  51673.64  11471.55  82498 	
26610	 	0  0  43661.37  0  43661.37 	
26611	 	13023.64  2619.7  44534.85  11638.94  71817.12 	
26613	 	13008.67  2881.67  49688.17  12918.1  78496.6 	
26615	 	12251.2  3112.2  43719  15363.4  74445.8 	
26617	 	13832  2881.67  52693.33  12802.83  82209.83 	
26618	 	14573  2377.38  44614.38  11763.38  73328.13 	
26621	 	13008.67  2881.67  52281.67  15799.77  83971.77 	
26623	 	16796  2939.3  60045.7  24774.1  104555.1 	
26625	 	16842.14  2375  51435.71  12024.29  82677.14 	
26628	 	14573  2377.38  39742.3  11763.38  68456.05 	
26631	 	16796  2401.39  67033.06  12061.83  98292.28 	
26634	 	16796  2161.25  66690  12380.88  98028.13 	
26636	 	14980.39  3143.64  49720.78  9905.66  77750.47 	
26884	 	11323.58  3293.33  50772.22  4147.77  69536.9 	
26888	 	11176.75  3087.5  49400  4145.48  67809.73 	
26892	 	11281.31  2964  48782.5  4146.31  67174.12 	
26896	 	11297.61  3163.82  43710.67  4146.27  62318.38 	
26900	 	11281.31  2964  49400  4146.31  67791.62 	
26904	 	11370.3  3113.45  49815.13  442.11  64740.98 	
26936	 	12751.38  2699.36  64925.71  9368.36  89744.8 	
26937	 	13187.06  2634.67  65866.67  5516.33  87204.72 	
26939	 	12725.51  2703.65  68425.68  9963.45  93818.28 	
26940	 	12839.88  2717  67184  9830.6  92571.48 	
26942	 	12813.13  2593.5  66690  12226.5  94323.13 	
26943	 	13056.99  257739.1  68014.49  11813.04  350623.7 	
26944	 	12870.46  2540.57  64572.86  11644.29  91628.18 	
26946	 	12653.14  2559.82  63321.82  11856  90390.77 	
26947	 	12813.13  2662.97  68696.88  12697.34  96870.31 	
26948	 	12587.5  2660  159663.3  12540  187450.8 	
27027	 	7883.34  4149.6  45497.4  19215.7  76746.04 	
27028	 	12328.59  4150.92  39717.6  18998.42  75195.53 	
27029	 	12325.3  4150.59  48646.65  20068.75  85191.29 	
27030	 	13529.01  4215.47  29640  21409.96  68794.44 	
27031	 	15.25  5.13  56.48  24.39  101.24

24                  	Compendium  of  Project  Reports	 	
 	
27032	 	12325.3  4150.42  45571.5  16816.58  78863.81 	
27033	 	7879.3  4149.6  45523.09  23427.95  80979.94 	
27034	 	12316.41  4150.39  45685.12  20069.24  82221.16 	
27035	 	12317.5  4150.47  43528.33  16849.08  76845.38 	
27036	 	12340.12  4150.26  46156.07  18093.57  80740.02 	
27037	 	12312.95  4150.42  45633.25  18394.64  80491.26 	
27038	 	7879.3  4149.6  45531.98  20068.75  77629.63 	
27039	 	12325.3  4150.42  58724.25  16816.58  92016.56 	
27040	 	12310.48  4150.48  45667.56  19244.59  81373.11 	
27041	 	12331.65  4150.45  45673.83  18516.53  80672.46 	
27087	 	13235.47  2990.41  39613.21  776.73  56615.82 	
27481	 	7125  3087.5  82792.5  1425  94430 	
27482	 	7410  2593.5  78299  1482  89784.5 	
27483	 	6062.73  2470  69204.91  8577.64  86315.27 	
27484	 	8027.5  2624.38  84937.13  1235  96824 	
27485	 	9115.48  2646.43  83744.76  17936.9  113443.6 	
27486	 	8468.57  2822.86  60956.07  19760  92007.5 	
27487	 	8336.25  2531.75  72865  14523.6  98256.6 	
27488	 	10034.38  3087.5  82204.69  1543.75  96870.31 	
27489	 	8159.82  2536.16  84906.25  21171.43  116773.7 	
27545	 	4350.39  2470  92505.48  14509.26  113835.1 	
27546	 	4310.15  2470  56717.38  35024.6  98522.13 	
27547	 	6567.29  2470  95203.97  35023.15  139264.4 	
27548	 	25137.54  2470  66460.64  37671.03  131739.2 	
27549	 	4322.5  2470  55155.1  35024.6  96972.2 	
27550	 	13290.95  2822.86  43166.19  37638.1  96918.1 	
27551	 	4310.39  2470  54654.8  34827  96262.2 	
27552	 	4293.31  2470  51448.98  35029.09  93241.38 	
27553	 	4314.78  2470  93242.5  35390.47  135417.8 	
27554	 	4324.4  4940  64752  35024.6  109041 	
27555	 	4878.25  2470  58230.25  35765.6  101344.1 	
27556	 	4322.5  2470  54932.8  35024.6  96749.9 	
27557	 	4310.15  2470  52034.67  35024.6  93839.42 	
27558	 	4318.24  2470  50784.05  35022.9  92595.19 	
27559	 	4314.78  2470  95017.81  35027.69  136830.3 	
27828	 	6029.4  790.4  23498.8  26260  56578.6 	
27836	 	7821.67  2161.25  16796  4116.67  30895.58 	
27837	 	11115  0  34394.75  6175  51684.75 	
27838	 	7014.8  0  24359.14  2470  33843.94 	
27839	 	7916.67  2216.67  28563.33  20583.33  59280 	
27840	 	6711.81  0  26037.09  1646.67  34395.57 	
27841	 	8710.87  988  20850.09  42283.11  72832.07 	
27842	 	5645.71  0  26287.86  44107.14  76040.71 	
27843	 	6962.31  1018.88  25826.94  3087.5  36895.63 	
27844	 	7014.8  988  31665.4  3087.5  42755.7 	
27845	 	9468.33  1097.78  30079.11  26072.22  66717.44

25 	
Compendium  of  Project  Reports 	
27846	 	4273.1  1729  38803.7  2470  47275.8 	
27847	 	8068.67  2223  16334.93  38696.67  65323.27 	
27848	 	6916  988  21406.67  2470  31780.67 	
27849	 	439.11  2401.39  11307.11  37050  51197.61 	
29626	 	9634.24  4273.1  17876.63  3099.85  34883.81 	
29627	 	9627.51  4259.38  16768.56  3098.48  33753.92 	
29628	 	9634.24  4273.1  16641.63  3096.76  33645.72 	
29629	 	9638.27  4268.16  11164.4  3099.03  28169.86 	
29630	 	9667.58  4281.33  17413.5  3103.97  34466.38 	
29631	 	9642.88  4262.51  18666.14  3102.32  35673.86 	
29632	 	9642.88  4262.51  16831.29  3098.09  33834.77 	
29633	 	9714.51  4248.4  31492.5  3097.38  48552.79 	
29634	 	9654.28  4271.2  15903  3098.14  32926.62 	
29635	 	9627.51  4259.38  13502.67  3098.48  30488.03 	
29636	 	9641.23  4281.33  70395  3099.85  87417.42 	
29637	 	9645.35  4281.33  55575  3098.48  72600.16 	
29638	 	9787.79  4215.47  50717.33  3108.91  67829.49 	
29639	 	9230.74  4234.29  67395.71  3098.09  83958.83 	
29640	 	9614.48  4199  72556.25  3118.38  89488.1 	
 
Tools and libraries used 
 	
ï¶	 numpy  	
ï¶	 matplotlib.pyplot  	
ï¶	 pandas  	
ï¶	 seaborn  	
ï¶	 sklearn  
 	
Data Processing and Result 
 	
Missing  data  and  null  values  were  treated.  Study  the  correlation  between  the 	
dependent  variable  (cost  A)  and  independent  variables  (human  labour,  seed,  fertilizer  and 
other expenses). For that we use heatmap. From the heatmap it is  clear that the dependent 
variable and independent variables are correlated.

26                  	Compendium  of  Project  Reports	 	
 	
Split  the  data  randomly  in  to  train  (80%) and  test  (20%).Using  the  train  data,  train 	
the  machine  and  construct  a  model  .Fit  the  test  value  in  the  model  and  we  got  the  future 
predicted  value.  Then  calculate  the  mean  square  error  and  regression  coefficients	.         	
Mse =0.004932663589052374 	and the regression coefficients are 	0.99999989	, 1.00000004, 	
1.00000001, 1.00000003.	 It is evident that the model that we construct is correct and using 	
it we can predict the future value. 	
 	
Actual and Predicted Values 	
 	y_test	 	y_pred	 	
327 	67653.30 	67653.298573 	
233 	22847.50 	22847.498535 	
122 	66126.20 	66126.189426 	
102 	41984.15 	41984.148164 	
71 	34412.19 	34412.188575 	
253 	81678.78 	81678.789970 	
12 	30109.30 	30109.298738 	
96 	33894.25 	33894.248445 	
330 	23629.26 	23629.258646 	
261 	46979.40 	46979.399467 	
328 	34130.46 	34130.458787 	
  90 	60223.40 	60223.408733 	
 
 	
 	
 
 	
To  check  the  accuracy  we  draw  scatter  diagram  with  actual  value  and  predicted 	
value.  From  the  diagram  it  is  clear  that  almost  all  the  values  lying  in  a  straight  line.  It 
means that actual value and predicted values are almost equal and the model that we create 
is a perfect one.

27 	
Compendium  of  Project  Reports 	
 	
 
 
Conclusion 
 	
It is aimed to discuss the application of artificial intelligence in agriculture, even in 	
the common case of small available datasets. The approach was applied and verified in the 
specific case of paddy. Also aims to use the model to predict other important crops. 
 
 
REFERENCE 
 	
1.     Report on Cost of Cultivation 2011-12 to 2020-21 published by Directorate of      Eonomics and 	
Statistics 	
2.    Multilinear regression model based studies

28                  	Compendium  of  Project  Reports

29 	
Compendium  of  Project  Reports 	
Forecasting of Cement Price in Kerala	
 	
Submitted  By 	
Sri.Sijith K.S., 	Research Officer

30                  	Compendium  of  Project  Reports	 	
 	
Introduction of the project 
          The  construction  sector  is  a  key  contributor  to  the  economy  of  Kerala,  with  cement 
being  a  vital  input  for  construction  activities.  The  price  of  cement,  have  a  significant  impact 
on the profitability and competitiveness of construction firms, as well as the affordability and 
accessibility  of  housing  for  the  general  public.  The  price  of  cement  in  the  state  is  subject  to 
fluctuations  due  to  various  factors,  including  changes  in  demand,  supply  constraints,  and 
economic  conditions.  Therefore,  accurate  forecasting  of  cement  prices  is  essential  for 
stakeholders  in  the  construction  industry,  including  cement  manufacturers,  distributors, 
retailers,  contractors,  and  policymakers.  Such  forecasting  can  help  these  stakeholders  make 
informed  decisions  regarding  production,  procurement,  pricing,  and  investment.  The 
motivation  of  this  project  is  to  develop  and  test  a  machine  learning-based  approach  for 
forecasting the price of cement in Kerala, using historical data on cement price 
             Accurate  forecasting  of  cement  prices  can  be  crucial  in  resolving  disputes  between 
contractors  and  house  owners  in  the  construction  industry.  When  contractors  agree  to 
construct  a  building  for  a  fixed  price,  they  base  their  calculations  on  the  estimated  costs  of 
materials,  including  cement.  If  the  price  of  cement  increases  unexpectedly  during  the 
construction process, the contractor may face difficulties in meeting their obligations and may 
seek  to  pass  on  the  additional  costs  to  the  house  owner.  On  the  other  hand,  if  the  price  of 
cement  decreases,  the  contractor  may  benefit  from  the  cost  savings  without  passing  them  on 
to the house owner. This can lead to disputes over payment, quality, and timelines. 
               By  providing  reliable  forecasts  of  cement  prices,  stakeholders  in  the  construction 
industry can mitigate the risks of such disputes. Contractors can use the forecasts to negotiate 
more  accurate  and  transparent  contracts  with  house  owners,  based  on  realistic  estimates  of 
material costs. House owners can use the forecasts to monitor the progress of construction, to 
ensure  that  the  quality  and  quantity  of  materials  used  are  consistent  with  the  agreed  terms. 
Additionally,  policymakers  can  use  the  forecasts  to  identify  potential  bottlenecks  in  the 
supply  chain  and  to  implement  measures  to  mitigate  the  impact  of  price  fluctuations  on  the 
construction  industry  and  the  wider  economy.  In  summary,  accurate  forecasting  of  cement 
prices can help to reduce the uncertainties and risks associated with construction projects, and 
can facilitate smoother and more efficient transactions between contractors and house owners. 
 Objectives of the project	 	
 
 	The main objective of the study is to forecast the price of cement in construction sector of 	
Kerala  using  historical  data  of  cement  prices 	through  machine  learning  based  approach.  The 	
purpose  is  to  help  the  construction  firm,  building  developers  and  householders  to  predict 
cement  prices  for  their  upcoming  projects  and  hence  construction  cost  of  the  project.  The 
rising  cost  of  building  materials  is  also  responsible  for  time  overrun  experienced  on  most 
construction  projects.  Another  impact  is  the  high  dispute  rate  between  contractor  and  owner 
due  to  the  rising  cost  of  building  materials.  In  this  study  machine  learning  algorithms,  may 
predict the changes in cement prices and forecast the possible prices. So the main objective is	 	
to develop a reliable model for forecasting the prices of cement in Kerala for the next 2 years.

31 	
Compendium  of  Project  Reports 	
Literature review  
      A  study  named  as  âApplication  of  machine  learning  in  cement  price  prediction  through  a   
web  based  systemâ,  has  conducted  by  A.O  Afolabi,  College  of  Science  and  Technology, 
Covenant  University,  Ota,  Nigeria.	  This    study    aims    to    predict    cement    prices    in    the  	
Nigerian    construction    industry    through    a    machine  learning  based    approach.  The  study 
aimed  to  develop  a  web  based  learning  platform  that  uses  machine  learning  algorithms  on 
historical data of cement prices, petrol prices, diesel prices, interest rate, and exchange rate to 
predict  future  prices  of  cement  products.  The  web  based  learning  platform  was  developed 
using  HTML,  CSS,  MySQL,  and  PHP.  For  building  a  reliable  machine  learning  model, 
python language was used to train the system with the available data. 
      There  is  another  study  conducted  by  olga  vainer  finance  director  at  ilanot  tel  aviv	 israel	 	
named  as  âmonthly  sales  of  french  champagne  -  arima  modelsâ. 	The  aim  of  the  study  was  to 	
predict  the  number  of  monthly  sales  of  champagne  for  the  Perrin  Freres  label  (named  for  a 
region  in  France).The  dataset  provides  the  number  of  monthly  sales  of  champagne  from 
January  1964  to  September  1972.The  validation  dataset  is  about  11%  of  the  original  dataset. 
ARIMA model was developed by selecting the appropriate parameters, including the order of 
differencing  (d),  the  autoregressive  order  (p),  and  the  moving  average  order  (q).  The  model 
was  then  validated  using  residual  analysis,  and  the  forecast  accuracy  was  evaluated  using 
RMSE. 
 
Data set used  
         Labour and  housing  section  of  Department  of  economics  and  statistics  collects  price  of 
building  materials  since  1974-75.  The  following  is  the price  of cement  in  Kerala  from  1974-
75 to 2021-22 and is the annual average prices of cement in all districts. 
 	
year  price of cement 	
in Rs for 50kg 	
year  price of cement 	
in Rs for 50kg 	
1974-75  17  1998-99  169 
1975-76  20  1999-00  169 
1976-77  20  2000-01  180 
1977-78  21  2001-02  182 
1978-79  25  2002-03  149 
1979-80  28  2003-04  167 
1980-81  32  2004-05  169 
1981-82  35  2005-06  185 
1982-83  48  2006-07  221 
1983-84  59  2007-08  250 
1984-85  70  2008-09  279 
1985-86  80  2009-10  268 
1986-87  80  2010-11  289 
1987-88  85  2011-12  336 
1988-89  90  2012-13  367 
1989-90  94  2013-14  354 
1990-91  100  2014-15  396 
1991-92  107  2015-16  422 
1992-93  120  2016-17  415 
1993-94  129  2017-18  395 
1994-95  140  2018-19  397 
1995-96  162  2019-20  393 
1996-97  178  2020-21  425 
1997-98  177  2021-22  428

32                  	Compendium  of  Project  Reports	 	
 	
Plot of cement price per 50 kg in Kerala from 1974-75 to 2021-22 	
 	
 	
 
Method and Methodology   
The following methodology was adopted in the study: 
 	
1.  Collect data on cement prices from reliable sources government publications, or 	
relevant websites 	
2.   Data cleaning and pre processing 
3.   Divide the data into training data set and test data set. 
4.  Component analysis of the time series to examine the trends, seasonality, and 	
other patterns in the data. 	
5.   Choose appropriate time series forecasting model. 
6.   Estimate the model parameters by fitting the chosen model to the time series 	
data. 	
7.   Train the chosen model using the training data 
8.   Evaluate the performance of the model on the test data. 
9.   Check the accuracy of the model 
10.  Check the adequacy and validity of the model through residual analysis and 	
conducting statistical tests. 	
11. Use the fitted model to make future predictions of cement prices. 
 	
        The  study  has  done  using  python  machine  language.  The  data  has  split  into  training 
dataset and testing dataset. The training data set  contains price of cement from  1974-75 to 
2017-18  and  testing  data  set  contains  data from  2018-19  to  2021-22.  We divide  a data  set 
into  a  training  set  and  a  testing  set  in  order  to  evaluate  the  performance  of  a  predictive 
model.  The  training  set  is  used  to  fit  or  train  the  model,  which  involves  estimating  the 
model  parameters  using  the  data  available.  The  model  is  then  tested  on  the  testing  set, 
which  is  a  subset  of  the  data  set  that  was  not  used  in  the  training  process.  The  purpose of 
using  a  separate  testing  set  is  to  assess  how  well  the  model  will  perform  on  new,  unseen 
data.  By  evaluating  the  model  on  the  testing  set,  we  can  estimate  the  model's  ability  to 
generalize to new data. If the model performs well on the testing set, it is likely to perform 
well on new, unseen data.

33 	
Compendium  of  Project  Reports 
 
       Component  analysis  is  a  technique  used  in  time  series  forecasting  to  break  down  a 
time  series  into  its  underlying  components,  which  include  trend,  seasonality,  and  noise  or 
error.  Component  analysis  is  a  powerful  tool  in  time  series  forecasting  that  can  help  us 
understand the underlying behaviour of a time series and make more accurate predictions. 
By separating out the different components, we can better capture the trends, patterns, and 
seasonality  in  the  data  and  reduce  the  impact  of  noise  or  error.On  conducting  component 
analysis  of  our  time  series  we  can  understand  that  our  time  series  has  a  trend  but  no 
seasonality and noise. So AR or ARIMA model time series forecasting is suitable.  	
 	
       	
 Also,  there  is  evidence  of  autocorrelation  in  the  data.  ARIMA  models  are 	
commonly used for time series forecasting when there is evidence of autocorrelation in the 
data. Autocorrelation refers to the dependence of a time series on its past values.

34                  	Compendium  of  Project  Reports	 	
 
     ARIMA  (Auto  Regressive  Integrated  Moving  Average)  is  a  popular  time  series 
modeling  technique  used  for  forecasting  future  values  based  on  past  observations  and 
fitting  errors.  It  is  a  combination  of  two  models:  the  autoregressive  (AR)  model  and  the 
moving average (MA) model. There are 3 parameters p, d, q in an ARIMA model. 
 
p-order of autoregressive terms 
d-number of differences done to make data stationary 
q-order of moving average terms 
   Stationarizing  a  time  series  is  a  common  technique  used  in  time  series  analysis  to 
simplify  and  improve  the  accuracy  of  statistical  models,  facilitate  interpretation  and 
forecasting  and  reduce  variability.  A  stationary  time  series  is  one  whose  statistical 
properties  such  as  mean,  variance,  autocorrelation,  etc.  are  constant  over  time.  ADF 
(Augmented Dickey-Fuller) test is a statistical test used to determine whether a time series 
is stationary or not. The null hypothesis of the test is that time series is non-stationery and 
alternative hypothesis is  that the series is stationary.	 If the probability of the test statistic; 	
p  value  <0.05(chosen  significance  level),  null  hypothesis  can  be  rejected  i.e.;  time  series 
will be stationery. In the case of our time series p-value = 0.998>0.05, indicates that time 
series is non-stationary. 
    The  technique  of  differencing  can  be  used  to  makes  the  data  stationary.  After  first 
differencing, we conducted Augmented Dicky fuller test and got the p-value = 0.85, which 
is  greater  than  0.05,  indicating it is non-stationary.  After  Second  differencing,  we  again 
conducted  Augmented  Dicky  fuller  test  and  got,  p-value  =  0.004,  which  is  less  than  0.05 
indicating now it is stationary.  	
       Here  we  had  done  two  times  differencing  to  stationarize  the  data.  So  the  value  of 	
parameter  d  of  the  ARIMA  model  can  be  taken  as  2.  Library  Pmdarima  in  python  has  an 
Auto_arima function, using this we can find possible ARIMA models with different orders 
and  corresponding  AIC  values.In  general,  a  lower  AIC  value  indicates  a  better  model  fit. 
(AIC  stands  for  Akaike  Information  Criterion,  which  is  a  statistical  measure  used  to 
compare the relative quality of different statistical models).In our study we got best model 
as ARIMA (3,2,0). Fitting of model was done using Python in test data set. Following plot 
is the fitted price (red color) and actual price (blue color) of cement.

35 	
Compendium  of  Project  Reports 	
After fitting the ARIMA model to the data, the next  step is to evaluate the model's 	
performance.  The  accuracy  of  the  model  can  be  checked  on  test  data  by  comparing  the 
predicted values from the model to the actual values in the test set. MAPE (Mean Absolute 
Percentage  Error)  is  a  commonly  used  measure  for  evaluating  the  accuracy  of  a  model.  It 
measures  the  average  percentage  difference  between  the  forecasted  values  and  the  actual 
values, and is expressed as a percentage. The formula for calculating MAPE is: 	
MAPE = (1/n) * â(|Actual - Forecast| / Actual) * 100 
Where, n is the number of observations in the data set. In our case MAPE is 3.87, it 	
means  that  on  average,  the  forecasted  values  were  off  by  3.87%  from  the  actual  values. 
And  R  Sq=  0.54,  it  means  that  54%  of  the  variability  observed  in  the  price  of  cement  is 
explained by the model.	  	
Finally  we  check  the  adequacy  of  the  model.  Residual  analysis  is  an  important 	
technique for  evaluating  the adequacy of time series forecasting models. Residuals are the 
differences between the actual values and predicted values. Some of the key conditions that 
should  be  satisfied  in  residual  analysis  are  residuals  should  be  normally  distributed  and 
residuals should not have any auto correlation. Q-Q plots can be used to check whether the 
residuals  are  normally  distributed  or  not.  If  the  residuals  fall  approximately  along  the 
straight line on the Q-Q plot, then the residuals are normally distributed. 	
 	
    	
The figure shows that residuals fall approximately  along a straight  line on the Q-Q 	
plot. So residuals are normally distributed. Shapiro-Wilkâs test is well known statistical test 
for  checking  the  normality.  The  null  hypothesis  of  the  Shapiro-Wilkâs  test  is  that  the 
population or sample follows a normal distribution.	 As by the test 	if the probability of the 	
test  statistic,  p-value  is  greater  than  chosen  significance  level  (0.05),  we  accept  the  null 
hypothesis.  In  our  case  it  was  0.58.so  we  accept  the  null  hypothesis  that  residuals  are 
normally distributed.

36                  	Compendium  of  Project  Reports	 	
 	
 	
     Next  we  have  to  check  whether  residuals  have  any  significant  auto  correlation  or  not. 
The  Autocorrelation  Function  (ACF)  plot  of  residuals  is  a  graphical  tool  used  to  diagnose 
the  presence of  autocorrelation  in  the residuals  of a  time  series  model.	 The  plot  shows  the 	
autocorrelation  on  the  y-axis  and  the  lag  on  the  x-axis.  If  the  autocorrelation  values  are 
within  the  confidence  intervals  (usually  the  blue  shaded  area  on  the  plot),  it  suggests  that 
there  is  no  significant  autocorrelation  in  the  residuals.  If  the  autocorrelation  values  fall 
outside  the  confidence  intervals,  it  indicates  that  there  is  significant  autocorrelation  and 
further investigation is needed. 
       	 	
From  the  ACF  plot  it  is clear  that  correlation  coefficients  between  residuals  at  any 	
lag except 0 are inside the blue band and hence statistically insignificant, means there is no 
significant auto correlation between residuals. So we can conclude that the chosen model is 
adequate.Hence  we  can  make  predictions  using  this  model.  Using  the  model  we  got  the 
forecasted price of cement for next two years as; 	
Year  Forecasted price of cement 
2022-23  429 
2023-24  437 	
   The results of this analysis suggest that the price  of cement is likely to increase by 
1-2% over the next two year.

37 	
Compendium  of  Project  Reports 	
 	
Tools and libraries used 
 	
ï¼	 	Pandas, Numpy- libraries used for data analysis and manipulation. 	
ï¼	 	Mathplotlib- libraries used for creating high-quality visualizations, including charts, 
plots, and graphs. 	
ï¼	 	Seaborn- data visualization library that is built on top of Matplotlib 	
ï¼	 	Statsmodels-library for statistical modeling and analysis 	
ï¼	 	Sklearn- library for machine learning 	
ï¼	 	Scipy-library for scientific computing and technical computing. 	
ï¼	 	Pmdarima- library for automatic time series forecasting with ARIMA models 	
 
Result 
        	The  aim of the study  was to develop a model that predicts the future price of cement 	
based on past prices of cement using python language. ARIMA(3,2,0) is the best model for 
the  data  and  that  predicts  the  average  price  of  cement  in  Kerala  for  the  next  2  years.  The 
results  of  our  analysis  suggest  that  the  price  of  cement  is  likely  to  increase  by  1-2%  over 
the  next  two  year.  However,  there  is  a  moderate  level  of  uncertainty  associated  with  the 
forecast  due  to  factors  such  as  fluctuations  in  raw  material  prices  and  changes  in 
government policies that could impact the cement market. This study helped to forecast the 
future price  of  cement  based  on  historical  data  at  reasonable accuracy  (MAPE=3.87).  The 
results indicated that the ARIMA model could effectively capture the patterns and trends in 
the  cement  price  data,  providing  accurate  and  reliable  forecasts  for  the  construction 
industry. 
Conclusion 
     Statistical model of the project provides a useful tool for predicting the prices of cement 
in Kerala, but it is important to keep in mind the potential uncertainties and risks associated 
with  any  forecasting  model.	 One  of  the  limitations  of  the  study  was  here  only  considered 	
the past prices of cement for forecasting the future prices. But there are some other factors 
that  affect  the  price  of  the  cement,  (eg.  petrol  price,  govt.policies  &  regulations,  currency 
exchange  rate  etc.)  which  are  not  considered  in  the  study.  However  this  study  helped  to 
forecast  the  future  price  of  cement  with  a  reasonable  accuracy. 	For  future  research,  other 	
prediction  model  can  be  developed  to  consider  other  building  materials such as steel 
reinforcement, bricks,  paints,  timbers etc. However,  it  should  be  noted  that the availability 
of  data may  be  impact the  model's  performance  and  the  model  may  need  to  be  updated  to 
maintain its accuracy.	It can conclude that price of any building materials and hence project 	
cost  can  be  forecast  with  a  reasonable  accuracy  using  python  and  machine  learning 
approach.  The  study  recommends  the  increased  use  of  machine learning  in  predicting 
building materials prices to aid in project cost estimation. 
 
References 	
 1. Monthly Sales of French Champagne - ARIMA Models, Olga Vainer Finance Director at Ilanot Tel 	
Aviv-Yafo, Tel Aviv District, Israel 	
2.  Application of machine learning in cement price prediction through a   web based system, by A.O 	
Afolabi, Oluwamayowa Abimbola Department of Building Technology, College of Science and 
Technology, Covenant University, Ota, Nigeria 	
3 .Data of Economics and statistics department Kerala

38                  	Compendium  of  Project  Reports

39 	
Compendium  of  Project  Reports 	
Index of Industrial Production 	
-An Overview	 	
Submitted  By 	
Smt. Sudarsha R., 	
Deputy Director	
 	
and	 Smt. Brijila R., 	
Statistical Assistant Grade II

40                  	Compendium  of  Project  Reports	 	
 
1. 	Introduction of the project 	
2. Objectives of the project 
3. Methodology and method used 
4. Datasets used 
5. Tools and libraries used 
6. Result with inference 
7. Conclusion 
8. Reference 
 
Contents 
 
SYNOPSIS on Index of Industrial Production (IIP)   
1 IIP- a perspective   
2.Introduction   
IIP (Index of Industrial Production)   
METHODOLOGY   
Selection of Item Basket   
Source of Data 
Weighting Diagram   
3. Objective/Motivation/need/scope of the project   
4.Literature survey   
Data processing-imputed the missing values   
The pictorial representation of the data   
Pictorial representation of Q1  
ARIMA MODEL   
Residual   
Density Plot of residual   
Prediction   
Analysisâ¦   
Similar models may be fitted to Q2,Q3 ,Q4 datasets  
Q2 Best model: ARIMA(0,1,0)   
Q3 Best model: ARIMA(2,1,1)   
Q4 Best model: ARIMA(3,1,0)   
Graph of QGen   
ARIMA MODEL   
Best model: ARIMA(3,1,0)   
Residual   
Density of residual   
Prediction   
Model 2: Support Vector Regression Approach   
Pictorial reprsnâ¦   
Real Values Vs Predicted values (kernel=âlinear)   
5.Proposed methodology:   
6. Expected outcome   
7. Conclusion:

41 	
Compendium  of  Project  Reports 	
 	
IIP (Index of Industrial Production) 	
1. Introduction  
                     	The  Index  of  Industrial  Production  (IIP)  is  an  index  that  shows  the growth 	
rates in different industry groups of the economy in a fixed period of time. 	It is a composite 	
indicator of the general level of industrial activity in the economy.  In India, it is calculated 
and published by the 	Central Statistical Organisation	 (CSO) every month. 	
                   IIP  measures  the  changes  in  the  industrial  production  and  the  general  level 
of  industrial  performance  in  the  economy.  It  is  used  as  a  short  term  macro-economic 
indicator  of  industrial  growth.  IIP  is  considered  to  be  one  of  the  lead  indicators  for  short 
term economic analysis because of its strong relationship with economic fluctuations in the 
economy.  It  is  also  used  as  major  data  source  for  the  compilation  of  annual  and  quarterly 
estimation and forecasts of GDP. Index of industrial production is compiled for three major 
sectors. 	
a.	 	Manufacturing 	
b.	 	Mining and Quarrying 	
c.	 	Electricity 	
 
2. Objectives  
 	
India  being  an  Agrarian  based  Country  in  the  pre  independence  period, 	
Government  was  compelled  to  focus  in  Manufacturing  sector  so  as  to  attain  economic 
development and  adapting most modern technologies in Agriculture. Hence, the  Industrial 
development began immediately after independence. IIP is a measure to assess the same. 
  	A  trend  in  industrial  sector  is  to  be  analysed  and  growth  in  each  sector  is  to  be 	
evaluated  so  that  policy  makers  can  address  the  issues  faced  by  the  industrial  production 
sector  and  support  with  appropriate  infrastructure  both  physical  as  well  as  financial. 
Government  already  have  taken  more  initiatives  to  promote  the  sector  by  easing  he 
licensing procedure, etcâ¦ 
 
3. Methodology 	 	
 
  	In  case  of  âmanufacturingâ  sector,  the  base  used  for  selection  of  items  in  the  item 	
basket  for  IIP  is  the  item-wise  production  data  of  Annual  Survey  of  Industries  (ASI),  
2011-12  provided  by  Central  Statistical  Organization  (CSO).    The  item  basket  for 
manufacturing  sector  was  identified  with  base  year  2011-12.    A  total  of 	23 	NIC  2-  digit 	
level  item  groups  were  selected  for  manufacturing  sector.    Since  the  number  of  items  in 
âmining  & quarryingâ sector in Kerala is very  few, all the mining items in the item basket 
are covered. The Electricity sector consists of single item and total electricity generated in 
Kerala is taken.  
Weighing Diagram

42                  	Compendium  of  Project  Reports	 	
 	
The  relative  importance  of  various  economic  activities  is  different  and  these 	
differentials  need  to  be  reflected  while  measuring  the  performance  of  the  entire  industrial 
sector.    With  a  view  to  achieving  this,  each  item  included  in  the  item  basket  is  given 
appropriate weight.  	
Total  weight,  which  is  taken  as  1000,  was  apportioned  first  to  three  industrial 	
sectors,  i.e.,  Manufacturing,  Mining  &  Quarrying  and  Electricity.    The  sectoral  weight  of 
manufacturing sector was then allocated to its 2 & 3- digit industry groups in proportion to 
their GVA figures.  Finally, the weight allocation at item level has been done in proportion 
to item-wise GVO. In case of mining & quarrying sector the total weight of the sector was 
apportioned  to  the  items  included  in  the  item  basket  in  proportion  to  their  value  of  output 
furnished by IBM. 	
 	
4. Datasets used  	
The  monthly  production  data  of  âmining  &  quarryingâ  and  âelectricityâ  for 	
compiling  State  level  IIP  are  collected  from  Indian  Bureau  of  Mines  (IBM),  Nagpur  and 
Central  Electricity  Authority  respectively.    In  case  of  âmanufacturingâ  sector,  after 
finalization  of  item  basket  (containing  selected  item  groups  and  items  within  the  group), 
item  wise  list  of  factories  along  with  their  production  in  the  base  year  has  been  prepared 
from  out  of  the  list  of  factories  from  ASI  survey  2011-12.  A  total  of  190  registered	 	
manufacturing  units  are  selected  for  data  collection  and  the  department  collects  monthly 
production data of 134 manufacturing items. 
 
   	The  data  source  is  from  the  office  of  Directorate  of  Economics  and  Statistics, 	
Government of Kerala. 
 
Data used is as follows,

43 	
Compendium  of  Project  Reports 	
 	
 	
Data processing-imputed the missing values 	
 	
 	
 
 
5.  Tools and libraries used 
 	
import pandas as pd	 	
import numpy as np 
from statsmodels.tsa.stattools import adfuller 
from pmdarima import auto_arima 
from pandas import read_csv 
from pandas import datetime 
from matplotlib import pyplot 
from statsmodels.tsa.arima.model import ARIMA 
from sklearn.metrics import mean_squared_error 
from math import sqrt

44                  	Compendium  of  Project  Reports	 	
 
6. Result with inference 	
The pictorial representation of the data	 	
 	
 	
Method 1:     To predict the first Quartile Q1 using ARIMA Model 	
Pictorial representation of Q1 	
 	
Check the Stationarity using Augmented Dickey-Fuller test 
 
 
 
 
 
 
 
 
 
 
 
 
Here the P-Value is greater than 0.05, the data is non-stationary. 
Being non- stationary, took first difference and on testing stationarity is found. 	
1.	 	ADF Statistic: 	-	1.8603658775718286 	 	
2.	 	P	-	Value: 0.3509296637282311 	 	
3.	 	n_lags: 0 	 	
4.	 	Num of observations used for ADF Regression and Critical Values   	 	
            Calculation: 40 	 	
5.	 	Critical Values:	 	
 1%, 	-	3.6055648906249997 	 	
 5%, 	-	2.937069375 	 	
 	 	10%, 	-	2.6	06985625

45 	
Compendium  of  Project  Reports 
                	
Graph of Q1& First difference (D2) 	
 	
 	
ïµ	 1.	ADF Statistic: -6.437777441920769  	
ïµ	 2.P-Value: 1.6365132616833867e-08  	
ïµ	 3.n_lags: 0  	
ïµ	 4.Number of observations used for ADF Regression and Critical Values 
Calculation : 39  	
ïµ	 5.Critial Values:  	
    1%, -3.610399601308181  
    5%, -2.939108945868946  
    10%, -2.6080629651545038 
                                                             
It is Stationary 
Using Auto ARIMA MODEL 	
ïµ	 Best model: ARIMA(1,1,0) 
For verifying the best model, find its Residual and density graph of Residuals 	
 	
Residual

46                  	Compendium  of  Project  Reports	 	
 	
Density Plot of residual	
 	
Prediction (Using the ARIMA model)	 	
 	
 
Method 2: Support Vector Regression Approach 
 	
ïµ	 Linear regression model is developed to forecast the Qgen using SVR  model. 	
ïµ	 Qgen is a linear combination of Q1,Q2,Q3 and Q4. 	
ïµ	 Data is plotted against each Qi with Gen Q as follows:

47 	
Compendium  of  Project  Reports 
Pictorial representation 	
 	
 
 	
ïµ	 Split the data in to test /train as  	
ïµ	  X = df[['Q1','Q2','Q3','Q4']] 	
ïµ	 y = df['QGenâ] 	
ïµ	 Support vector regressor with linear kernel is chosen with test score as 0.9832  	
ïµ	 Support vector regressor with Gaussian(radial basis function) kernel is also done 
with a test score of 0.03352. 	
ïµ	 RMSE for linear SVR: 10.9163 RMSE for RBF kernelized SVR: 82.9108 	
Real Values Vs Predicted values (kernel=âlinear)

48                  	Compendium  of  Project  Reports	 	
 	
 	
 Conclusion 	
ïµ	 	Trend has got irregular pattern. 	
ïµ	 	Tested for stationarity using Augmented Dickey-Fuller test. 	
ïµ	 	Being nonstationary, took first difference and on testing stationarity is found. 	
ïµ	 	ARIMA(1,1,0) the model suitable to the dataset. 	
ïµ	 	Support Vector Method is the best method for the prediction of this dataset. 	
ïµ	 	Analysis of prediction will be helpful for estimation of GDP. 	
 
 
 
 Reference 
 	
1.	 	Hand book from Central Statistical Organisation on IIP. 	
2.	 	Data from Directorate of Economics and Statistics, Government of Kerala. (Annexure-1) 	
3.	 	âSupport Vecor Regressionâ by Dr.Thirthjyoti Sarkar,Fremont,CA94536 	
4.	 	Machine Learning: Algorithms, Real-World Applications and Research Directions by Iqbal H Sarkar 
in âSN Computer Science2,â AN:160(2021) 	
5.	 	Review of ML and AutoML Solutions to Forecast Time-Series Data 	
6.	 	Ahmad Alsharef, Karan Aggarwal, Sonia, Manoj Kumar & Ashutosh Mishra inâ Archives of 
Computational Methods in Engineeringâ 29,2022 	
7.	 	ARIMA and SARIMA forecasting in Neptune.ai https://neptune.ai/blog/arima-sarima-real-world-
time-series-forecasting-guide#:~:text=forecasting%20have%20been-,some,-of%20the%20key

49 	
Compendium  of  Project  Reports 
 	
Prediction of Cost of Cultivation of 	
important Crops in Kerala	 	
A Machine Learning Approach	
 	
Submitted By 	
Smt. Suma S.A., 	
Research Assistant

50                  	Compendium  of  Project  Reports	 	
 
INTRODUCTION  
        	In  our  country,  agriculture  is  the  principal  pillar  of  the  economy.  The  majority  of 	
families  are  dependent  on  agriculture.  The  country's  GDP  is  primarily  focused  on 
agriculture.  More  than  half  of  the  land  is  used  for  agriculture  to  meet  the  needs  of  the 
population  of  the  region.  It  is  necessary  to  modernize  agricultural  practices  to  meet  the 
demanding requirements. In the past few years, a lot of fluctuation in the prices of the crop 
has  been  seen.  This  has  increased  the  rate  of  crop  damage  produced  each  year.  The  main 
aim of this prediction system is to ensure that the farmers get a better idea about the cost of 
cultivation and how to maximize the production. 
 
OBJECTIVE OF THE PROJECT 
 	
Cost  of  cultivation  of  major  crops  are  very  essential  for  formulating  proper  support 	
price policies, creating marketing facilities and assessing loss out of natural calamities, and 
the  share  of  agriculture  in  GDP.  We  need  reliable  data  on  crop  husbandry  right  from 
sowing to harvesting stages. For this, the department conducts an annual survey on Cost of 
Cultivation of Important Crops in Kerala during every agricultural year (July to June). The 
major  components  of  the  costs  involved  are  seed/seedlings,  fertilisers,  labour,  rent, 
equipment,  irrigation  charges  etc.The  main  objective  of  the  survey  is  to  produce  reliable 
estimates  on  production  cost  involved  in  major  agricultural  crops  in  the  State.  Major 
perennial,  annual  and  seasonal  crops  are  covered  in  the  survey  with  a  breakup  of  cost 
incurring during various stages from sowing to harvest.  
METHODOLOGY AND METHOD USED  
        The  survey  covered  all  the  districts  in  the  state  by  considering  taluk  as  a  stratum.  
From  each  Taluk,  required  numbers  of  investigator  zones  were  selected  using  circular 
systematic sampling method. From selected zones, cultivators and corresponding holdings 
are  selected.  The  holdings  are  grouped  under  three  size  classes  viz;  small,  medium  and 
large according to the area. 	. The investigators visited the selected holdings/cultivators and 	
collected the required information in the prescribed format. 
      Many  factors  are  influential  in  agriculture,  especially  in  cultivation  and  production.  In 
this work, the impact of different factors that affect the cost of cultivation is considered and 
predicts  how  these  factors  affect  the  cost  and  forecast  the  cost.  For  that  we  use  linear 
regression based prediction models. It is a machine learning algorithm based on supervised 
learning.  It  is  used  for  finding  out  the  relationship  between  variables  and  forecasting. 
Linear  regression  is  a  powerful  tool  for  understanding  and  predicting  the  behavior  of  a 
variable.  It  is  a  statistical  approach  for  modeling  the  relationship  between  a  dependent 
variable and a given set of independent variables.  
        Machine learning is a popular technique for finding patterns and linear and non-linear 
relationships  between  multiple  variables  (ML).From  the  static  point  of  view,  a  model  is 
counted  as  linear  if  the  modelâs  parameters  are  linear.  Classification,  Regression,  and 
Clustering  are  a  few  of  the  sub  categories  of  ML  that  can  be  utilized  for  analysis  and 
decision  making.  Machine  learning  is  becoming  more  and  more  popular  in  agriculture, 
where  it  is  usually  necessary  to  examine  complex  linkages  in  order  to  address  difficult 
agro-engineering difficulties.

51 	
Compendium  of  Project  Reports 
DATA SET USED 
 
         The actual cost incurred by the  cultivators for cultivating Tapioca  during 2019-20 is 
considered for the study. Cost incurred for human labour, seed, fertilizer and other 
expenses are taken into consideration. The data set have 469 rows and 5 columns .  
 	
Farmer_ID  human_labour  fertilizer  Other_Expences  seed  Total_cost 	
18690  34717.22222  12034.38889  28816.66667  5008.611111  80576.88889 
18692  45283.33333  11252.22222  35677.77778  6861.111111  99074.44444 
18694  62510  33820  0  7030  103360 
18696  50388  0  0  6051.5  56439.5 
18698  60515  11197.33333  0  6669  78381.33333 
18700  41397.2  13276.25  33345  4322.5  92340.95 
18702  54290.6  23450.18  0  4890.6  82631.38 
18703  49449.4  10991.5  27170  2667.6  90278.5 
18705  41835.625  11747.9375  33962.5  2315.625  89861.6875 
18707  44636.42857  8733.214286  8997.857143  2593.5  64961 
18709  40382.25455  13540.09091  24924.54545  3817.272727  82664.16364 
18711  64590.5  16479.01667  29228.33333  4462.466667  114760.3167 
18712  47053.5  14602.64  12597  4722.64  78975.78 
18713  35057.53333  11197.33333  24700  5705.7  76660.56667 
18714  39245.55556  10072.11111  41166.66667  5543.777778  96028.11111 
18956  105222  52827.125  1235  29640  188924.125 
18961  77064  38645.62  2470  0  118179.62 
18965  58390.8  17722.25  3705  29491.8  109309.85 
18966  28322.66667  17590.51667  2470  30875  79258.18333 
18967  87602.66667  564930.1667  3293.333333  30628  686454.1667 
18968  29640  18722.6  2470  29640  80472.6 
18969  62573.33333  28116.83333  4116.666667  30875  125681.8333 
18970  62320  18453.75  6175  29212.5  116161.25 
19094  28816.66667  38285  6861.111111  548.8888889  74511.66667 
19095  64064  26416  26000  520  117000 
19096  50684.4  41730.65  2470  555.75  95440.8 
19098  105057.3333  16466.66667  8233.333333  55.98666667  129813.32 
19130  161538  56810  3705  19760  241813 
19131  135603  47918  4940  29640  218101 
19257  130169  48988.33333  12350  19760  211267.3333 
19258  55328  40343.33333  6586.666667  26346.66667  128604.6667 
19259  58539  46930  9880  29640  144989 
19260  171171  45448  12350  29640  258609 
19261  82251  56711.2  7410  29640  176012.2 
19262  136344  50223.33333  28816.66667  29640  245024 
19263  166058.1  46930  2964  19760  235712.1 
19264  166725  38285  4940  19760  229710 
19265  171665  49400  4116.666667  19760  244941.6667 
19266  139900.8  106111.2  2964  29640  278616 
19267  152349.6  75088  3952  19760  251149.6

52                  	Compendium  of  Project  Reports	 	
 	
19268  156104  49400  32933.33333  29640  268077.3333 
19269  106704  37050  9880  29640  183274 
19661  108680  59280  4199  10003.5  182162.5 
19666  19760  20377.5  4199  10003.5  54340 
19669  30875  14820  4199  10003.5  59897.5 
19671  27170  14820  4199  10003.5  56192.5 
19676  96970.37037  14637.03704  4253.888889  10017.22222  125878.5185 
19680  70148  21859.5  4199  10028.2  106234.7 
19684  4940  22847.5  4199  9941.75  41928.25 
19687  82745  22847.5  4199  10003.5  119795 
19689  3458  14820  9151.35  7521.15  34950.5 
19693  113620  26552.5  4199  10003.5  154375 
19699  15561  14820  9151.35  10028.2  49560.55 
19702  21736  13832  4199  10028.2  49795.2 
19703  37050  15437.5  4199  10003.5  66690 
19704  20583.33333  14820  4199  10003.5  49605.83333 
19821  11263.2  16321.76  2964  13832  44380.96 
19822  65702  18278  3087.5  6792.5  93860 
19823  45052.8  17161.56  2223  11065.6  75502.96 
19824  136097  21538.4  2593.5  6792.5  167021.4 
19825  119671.5  20501  7039.5  13585  160797 
19963  50684.4  20871.5  16919.5  10868  99343.4 
19965  70202.88889  24420.06667  25660.55556  10977.77778  131261.2889 
19966  54182.81818  29943.13636  24363.18182  6848.636364  115337.7727 
19968  46930  22736.35  2161.25  6792.5  78620.1 
19969  78216.66667  12452.91667  2058.333333  6792.5  99520.41667 
19971  70082.13333  23802.56667  988  13585  108457.7 
19973  54193.25294  22814.80882  25353.82353  10926.11765  113288.0029 
19974  81658.2  58032.65  1605.5  6792.5  148088.85 
19976  62573.33333  63369.22222  1921.111111  10977.77778  138841.4444 
20030  42813.33333  33657.86667  27170  1235  104876.2 
20031  132741.9167  40343.33333  15272.83333  2964  191322.0833 
20032  0  33345  34580  4446  72371 
20033  46312.5  25276.33333  6463.166667  3499.166667  81551.16667 
20034  0  33345  31286.66667  4693  69324.66667 
20035  21736  25984.4  15561  2470  65751.4 
20036  64220  20741.66667  14566.66667  3166.666667  102695 
20037  21736  23415.6  14943.5  4075.5  64170.6 
20039  39520  13356.525  17290  4168.125  74334.65 
20040  110676.5833  20274.58333  40508  3952  175411.1667 
20042  63490.22727  17645.5303  17514.54545  3181.060606  101831.3636 
20043  106366.4333  18730.83333  20171.66667  4075.5  149344.4333 
20044  138715.2  38976.6  20583.33333  2840.5  201115.6333 
20396  50943.75  14820  3952  10497.5  80213.25 
20397  131733.3333  8562.666667  3211  10291.66667  153798.6667 
20398  87067.5  7113.6  27170  9262.5  130613.6

53 	
Compendium  of  Project  Reports 	
20399  83362.5  5804.5  2470  9262.5  100899.5 
20400  64837.5  12350  3396.25  10806.25  91390 
20742  46312.5  16252.6  1482  2223  66270.1 
20744  44305.625  19797.05  13214.5  2223  79540.175 
20754  27092.8125  19142.5  1482  2223  49940.3125 
20771  41852.77778  19691.38889  1482  2223  65249.16667 
20774  45108.375  15363.4  13214.5  1852.5  75538.775 
20778  65240.21739  17808.7  7844.934783  2223  93116.85217 
20781  42144.375  15561  864.5  1852.5  60422.375 
20826  61132.5  15437.5  4940  592.8  82102.8 
20829  48341.42857  12173.57143  4410.714286  564.5714286  65490.28571 
20830  50388  19809.4  27170  559.8666667  97927.26667 
20834  45882.12121  22649.15152  65118.18182  591.3030303  134240.7576 
20837  95977.14286  7410  0  555.75  103942.8929 
20839  31813.6  27861.6  2470  1037.4  63182.6 
20844  36414.85714  19834.1  3175.714286  1287.928571  60712.6 
20845  75852.90323  44101.45161  3983.870968  2788.709677  126726.9355 
20846  62604.11215  46456.7757  2885.514019  1442.757009  113389.1589 
20847  49297.08333  20531.875  2572.916667  568.1  72969.975 
20877  87771.45  11317.54  17537  2741.7  119367.69 
20878  121735.7143  11673.69048  14702.38095  2858.142857  150969.9286 
20879  115892.4  12251.2  18772  2815.8  149731.4 
20880  88623.6  12238.85  7904  2741.7  111508.15 
20881  119301  11073.83333  14820  2717  147911.8333 
20882  122141.5  12257.375  14820  2964  152182.875 
20883  129400.5556  13630.74074  0  2881.666667  145912.963 
20884  129974.3939  11564.09091  14221.21212  2919.090909  158678.7879 
20885  88055.5  11621.35  14326  2758.99  116761.84 
20886  123088.3333  10209.33333  32933.33333  2791.1  169022.1 
20887  80599.1875  11701.625  27787.5  2732.4375  122820.75 
20888  80050.45455  10430.13636  20209.09091  2761.909091  113451.5909 
20889  85783.1  12379.64  20254  2741.7  121158.44 
20890  130795.1163  11344.76744  14360.46512  2843.372093  159343.7209 
20891  121573.4  12201.8  18772  2781.22  155328.42 
20922  91595.83333  37616.04167  13379.16667  0  142591.0417 
20923  44460  34394.75  18525  0  97379.75 
20924  102505  36020.83333  23465  0  161990.8333 
20925  101270  33670.21667  27581.66667  0  162521.8833 
20926  105386.6667  35362.16667  24974.44444  0  165723.2778 
20927  49400  35609.16667  23670.83333  0  108680 
20928  104925.6  33846.41  27417  0  166189.01 
20929  69160  35537.125  15437.5  0  120134.625 
20930  69160  35490.8125  2778.75  0  107429.5625 
20932  96947.5  34541.40625  2624.375  0  134113.2813 
20933  99623.33333  34857.875  3087.5  0  137568.7083 
20934  99381.17647  35451.76471  2760.588235  0  137593.5294

54                  	Compendium  of  Project  Reports	 	
 	
20935  91390  34166.275  15437.5  0  140993.775 
20936  112693.75  34721.51042  3087.5  0  150502.7604 
21201  0  10571.6  2667.6  0  13239.2 
21203  19760  10374  2717  12350  45201 
21204  127205  8274.5  3705  11115  150299.5 
21205  127205  13585  4199  12350  157339 
21207  55081  7508.8  3754.4  9880  76224.2 
21210  32418.75  5829.2  27911  9880  76038.95 
21213  67153.125  8892  3087.5  10806.25  89938.875 
21463  40343.33333  118683.5  11320.83333  2470  172817.6667 
21464  150747.1875  25626.25  0  2470  178843.4375 
21466  86958.52941  10868  1452.941176  2411.882353  101691.3529 
21595  51870  10880.35  22230  2593.5  87573.85 
21596  22791.36364  9414.068182  785.9090909  2357.727273  35349.06818 
21598  21612.5  9639.175  988  2470  34709.675 
21600  51870  9945.866667  823.3333333  2634.666667  65273.86667 
21603  51870  11197.33333  823.3333333  2470  66360.66667 
21604  51870  11197.33333  2881.666667  2387.666667  68336.66667 
21605  40343.33333  11650.16667  9262.5  2470  63726 
21606  92393.4375  11933.1875  17135.625  2593.5  124055.75 
21607  46106.66667  9934.888889  4116.666667  2524.888889  62683.11111 
21608  24378.9  9361.3  52858  2568.8  89167 
21609  46106.66667  9934.888889  7410  2524.888889  65976.44444 
21796  255645  59280  3087.5  0  318012.5 
21797  228228  44460  4940  0  277628 
21798  213408  50635  4940  11856  280839 
21799  251940  50282.14286  8821.428571  10938.57143  321982.1429 
21800  205504  105798.3333  3293.333333  11856  326451.6667 
21801  281580  29640  6175  11362  328757 
21802  167960  83980  26758.33333  11362  290060.3333 
21803  196701.8182  70507.27273  2245.454545  10778.18182  280232.7273 
21804  222300  67925  4940  11362  306527 
21805  222892.8  70148  3952  11856  308848.8 
21806  247000  56810  4116.666667  11856  319782.6667 
21807  243048  50635  32110  11609  337402 
21820  41249  41496  4940  4100.2  91785.2 
21822  50388  43048.57143  4234.285714  4128.428571  101799.2857 
21823  51652.05882  42498.52941  3632.352941  4140.882353  101923.8235 
21825  0  35074  4940  3359.2  43373.2 
21826  0  40137.5  6175  3334.5  49647 
21828  82745  42731  9880  3359.2  138715.2 
21831  0  41743  9262.5  3211  54216.5 
22088  19595.33333  10209.33333  19760  22641.66667  72206.33333 
22089  27875.71429  11820.71429  42342.85714  22935.71429  104975 
22090  57760  18810  0  22800  99370 
22091  11945.81818  7320.181818  39295.45455  22454.54545  81016

55 	
Compendium  of  Project  Reports 	
22094  27664  4347.2  6175  17784  55970.2 
22096  29640  10868  0  21406.66667  61914.66667 
22097  29640  7410  5292.857143  15525.71429  57868.57143 
22100  27664  23218  0  22230  73112 
22101  31051.42857  10444.57143  4940  19407.14286  65843.14286 
22102  25688  10670.4  4940  22230  63528.4 
22397  27787.5  7718.75  9262.5  35506.25  80275 
22401  17100  9500  8550  35625  70775 
22411  32521.66667  10291.66667  6586.666667  46312.5  95712.5 
22412  29640  6175  8645  38593.75  83053.75 
22417  24700  10291.66667  9056.666667  43225  87273.33333 
22420  18525  15437.5  8645  38593.75  81201.25 
22704  79534  9880  3293.333333  5763.333333  98470.66667 
22708  98800  69160  2470  13832  184262 
22710  44954  16302  12350  5187  78793 
22712  82004  11856  988  13832  108680 
22714  60515  29516.5  7410  8645  106086.5 
22716  56480.66667  13585  11526.66667  5763.333333  87355.66667 
22717  77805  7619.95  3087.5  10806.25  99318.7 
22752  77805  6010.333333  0  1482  85297.33333 
22754  113414.1667  14820  20583.33333  2675.833333  151493.3333 
22755  104710.3571  9527.142857  17642.85714  2117.142857  133997.5 
22759  70395  10538.66667  0  2634.666667  83568.33333 
22761  109297.5  5038.8  0  1976  116312.3 
22762  67184  15017.6  0  1284.4  83486 
22763  42907.42857  11115  35285.71429  1270.285714  90578.42857 
22764  79781  8768.5  18525  2161.25  109235.75 
22765  88384.83333  14161.33333  12350  1893.666667  116789.8333 
22766  57489.25  8398  18525  1482  85894.25 
22924  66690  25688  17290  11115  120783 
22926  160550  45695  4940  17290  228475 
22927  146965  27417  19760  12967.5  207109.5 
22929  87067.5  23526.75  4013.75  12967.5  127575.5 
22931  81734.54545  24879.63636  15942.72727  19198.63636  141755.5455 
22932  138814  42879.2  11856  19760  213309.2 
22933  100198.6375  25163.125  15283.125  12350  152994.8875 
22934  0  103122.5  0  0  103122.5 
22935  107856.6667  21200.83333  13585  14820  157462.5 
22937  89414  25885.6  14326  9089.6  138715.2 
22938  0  18525  14820  21612.5  54957.5 
22951  0  22847.5  15746.25  12350  50943.75 
22952  213655  26552.5  16672.5  22230  279110 
22953  101270  25317.5  14820  17290  158697.5 
23427  7410  9213.1  6175  4446  27244.1 
23428  19760  11559.6  617.5  4199  36136.1 
23429  5903.3  15264.6  1432.6  3853.2  26453.7

56                  	Compendium  of  Project  Reports	 	
 	
23430  22649.9  11053.25  12547.6  3912.48  50163.23 
23431  20315.75  11003.85  3458  4446  39223.6 
23432  8398  22081.8  3952  3952  38383.8 
23433  4940  10308.13333  1646.666667  4083.733333  20978.53333 
23434  14820  16055  4446  4446  39767 
23435  17290  14523.6  3952  3754.4  39520 
23437  0  15906.8  4446  4693  25045.8 
23438  8645  22600.5  3211  3952  38408.5 
23439  0  22057.1  2717  4149.6  28923.7 
23441  6817.2  10275.2  2568.8  3754.4  23415.6 
23983  63016.863  49400  0  8645  121061.863 
23985  42011.242  37050  0  7410  86471.242 
23988  95319.62471  29058.82353  0  7410  131788.4482 
23990  52514.0525  0  0  7410  59924.0525 
24177  32933.33333  54718.73333  0  9880  97532.06667 
24179  24700  60515  0  9880  95095 
24183  18525  34147.75  617.5  9880  63170.25 
24185  54340  61354.8  0  9880  125574.8 
24187  28816.66667  38202.66667  1235  9880  78134.33333 
24189  25688  40014  1976  9880  77558 
24192  49400  62342.8  0  9880  121622.8 
24194  61750  70395  0  9880  142025 
24196  37050  45373.9  0  9880  92303.9 
24198  49400  57616.86667  0  9880  116896.8667 
24201  30875  38068.875  1235  9880  80058.875 
24203  37050  42138.2  2470  4940  86598.2 
24205  27444.44444  43515.91111  1097.777778  9880  81938.13333 
24207  36388.39286  44007.90179  1102.678571  9262.5  90761.47321 
24209  33592  41377.44  1235  9262.5  85466.94 
24706  69160  44460  741  9880  124241 
24707  69160  44460  0  19760  133380 
24708  74100  41990  1976  14820  132886 
24709  102999  49400  889.2  18525  171813.2 
24710  108927  46831.2  494  9880  166132.2 
24713  106704  49400  3309.8  11856  171269.8 
24715  103740  39520  1432.6  24700  169392.6 
24716  86450  31122  543.4  14820  132935.4 
24856  15808  30628  9880  2470  58786 
24857  52693.33333  34888.75  24700  8233.333333  120515.4167 
24858  67184  31369  17290  7904  123747 
24859  26346.66667  51046.66667  12350  4116.666667  93860 
24860  98800  40137.5  18525  7718.75  165181.25 
24861  93860  25729.16667  20583.33333  5763.333333  145935.8333 
24862  71136  56513.6  29640  7410  164699.6 
24863  22880  52260  20800  4550  100490 
24864  13173.33333  343988.6667  20583.33333  4528.333333  382273.6667

57 	
Compendium  of  Project  Reports 	
24865  29640  46189  18525  4322.5  98676.5 
24866  34580  35969.375  18525  9262.5  98336.875 
24867  72453.33333  25935  20583.33333  8233.333333  127205 
24868  92213.33333  32933.33333  16466.66667  8233.333333  149846.6667 
24869  19760  53599  24700  3705  101764 
24870  29640  45695  24700  5557.5  105592.5 
24902  44460  21964  1672  3420  71516 
24903  46106.66667  0  50058.66667  3705  99870.33333 
24904  43754.28571  21877.14286  1834.857143  4940  72406.28571 
24910  17784  0  2766.4  5928  26478.4 
24912  21171.42857  299.9285714  0  35.28571429  21506.64286 
24916  30875  30463.33333  0  10291.66667  71630 
24918  15808  23316.8  0  4940  44064.8 
24920  7904  1729  1778.4  19.76  11431.16 
24921  15437.5  14279.6875  694.6875  4631.25  35043.125 
24922  49811.66667  16672.5  1008.583333  3293.333333  70786.08333 
24923  18833.75  78249.6  1420.25  19760  118263.6 
24924  26111.42857  0  0  0  26111.42857 
25013  192660  0  34580  3705  230945 
25014  40508  5434  13832  3705  63479 
25015  62244  3260.4  6669  3705  75878.4 
25016  115596  4693  13832  2470  136591 
25018  53352  16450.2  15808  5928  91538.2 
25078  85956  7706.4  8151  3087.5  104900.9 
25080  44460  13585  22230  2470  82745 
25083  133380  7410  16055  2470  159315 
25084  87932  10291.66667  10703.33333  2470  111397 
25085  53352  4841.2  15808  1811.333333  75812.53333 
25086  65866.66667  14408.33333  6586.666667  2744.444444  89606.11111 
25087  22230  7718.75  21303.75  5557.5  56810 
25088  119795  0  14408.33333  5763.333333  139966.6667 
25090  44460  17784  6175  4940  73359 
25337  24206  10769.2  3705  2173.6  40853.8 
25338  13832  15923.26667  3705  2107.733333  35568 
25339  31122  172692.52  3705  2074.8  209594.32 
25341  34580  14326  3705  2124.2  54735.2 
25348  34580  10868  3705  2173.6  51326.6 
25350  11970  16739  28500  2128  59337 
25353  31122  9287.2  3705  2074.8  46189 
25355  34580  16845.4  3705  2124.2  57254.6 
25358  21612.5  20346.625  4322.5  2161.25  48442.875 
25364  17290  22946.3  4322.5  2161.25  46720.05 
25365  34580  18747.3  28405  2074.8  83807.1 
25452  81345.33333  14881.75  2470  1729  100426.0833 
25453  58381.81818  14881.75  2470  1729  77462.56818 
25454  53516.66667  14881.75  2470  1729  72597.41667

58                  	Compendium  of  Project  Reports	 	
 	
25455  61651.2  14881.75  2470  1729  80731.95 
25456  121459.5652  14881.75  2470  1729  140540.3152 
25457  58381.81818  14881.75  2470  1729  77462.56818 
25458  57923.92157  14881.75  12156.27451  1729  86690.94608 
25459  77064  14881.75  2470  1729  96144.75 
25460  58868.33333  14881.75  2470  1729  77949.08333 
25461  57798  14881.75  2470  1729  76878.75 
25462  72247.5  14881.75  2470  1729  91328.25 
25463  73050.25  14881.75  2470  1729  92131 
25464  86697  14881.75  14820  1729  118127.75 
25465  71926.4  14881.75  2470  1729  91007.15 
25466  62079.33333  14881.75  2470  1729  81160.08333 
25564  128440  35815  12350  7410  184015 
25568  104975  31863  12350  9880  159068 
25569  79040  16302  7410  5557.5  108309.5 
25571  115266.6667  33756.66667  10291.66667  4631.25  163946.25 
25572  100776  28652  4940  22230  156598 
25573  125146.6667  34580  8233.333333  5187  173147 
25574  209950  30875  6861.111111  4116.666667  251802.7778 
25575  126587.5  19945.25  33962.5  23156.25  203651.5 
25576  59280  0  12350  7286.5  78916.5 
25577  65866.66667  31904.16667  10291.66667  9365.416667  117427.9167 
25579  39520  27170  8645  12350  87685 
25580  44460  27170  8645  8398  88673 
25586  37050  41166.66667  0  10291.66667  88508.33333 
25869  57633.33333  13412.1  20583.33333  1976  93604.76667 
25870  69160  14375.4  0  1976  85511.4 
25871  60515  14227.2  12350  1976  89068.2 
25872  57633.33333  12720.5  0  1646.666667  72000.5 
25873  69160  13930.8  0  1976  85066.8 
25874  60515  13486.2  0  1976  75977.2 
25875  57633.33333  13634.4  0  1976  73243.73333 
25905  60515  12844  14408.33333  1976  89743.33333 
25906  57633.33333  13930.8  12350  1976  85890.13333 
25907  58209.66667  14227.2  13996.66667  1976  88409.53333 
25908  56192.5  12893.4  2470  1976  73531.9 
25909  69160  14375.4  0  1976  85511.4 
25910  55328  13930.8  12350  1976  83584.8 
25911  60515  13782.6  0  1976  76273.6 
25912  59280  14057.82857  0  1976  75313.82857 
26343  74100  25416.3  18590.86667  3077.62  121184.7867 
26345  74100  25218.7  18599.1  3008.46  120926.26 
26347  67925  26610.13333  3770.866667  3008.46  101314.46 
26350  74100  25218.7  18599.1  3008.46  120926.26 
26352  74100  25218.7  3779.1  3008.46  106106.26 
26354  83362.5  25218.7  18599.1  3008.46  130188.76

59 	
Compendium  of  Project  Reports 	
26356  74100  25218.7  3779.1  3008.46  106106.26 
26358  71250  25897  18582  3032.4  118761.4 
26360  74100  25539.8  3770.866667  3008.46  106419.1267 
26361  69468.75  25700.35  3766.75  3008.46  101944.31 
26364  69468.75  25700.35  3766.75  3008.46  101944.31 
26367  74100  25539.8  3770.866667  3008.46  106419.1267 
26370  64837.5  25539.8  18557.93333  2991.17  111926.4033 
26372  83362.5  25712.7  18599.1  3008.46  130682.76 
26373  72041.66667  27206.59259  3769.037037  2939.3  105956.5963 
26732  59280  10497.5  3087.5  5094.375  77959.375 
26733  79040  10127  4116.666667  4734.166667  98017.83333 
26734  70571.42857  10850.35714  4410.714286  4719.464286  90551.96429 
26735  72453.33333  10127  4116.666667  4693  91390 
26736  98800  10127  4116.666667  4693  117736.6667 
26737  108680  11609  3705  4723.875  128717.875 
26738  119600  11440  3900  4712.5  139652.5 
26739  79040  11115  4116.666667  4693  98964.66667 
26740  74100  10868  3705  4693  93366 
26741  81765.51724  11498.27586  4258.62069  4727.068966  102249.4828 
26742  26346.66667  11115  4116.666667  4693  46271.33333 
26743  63232  11312.6  4940  4693  84177.6 
26744  59280  10991.5  3705  4693  78669.5 
26745  69160  11732.5  3705  4693  89290.5 
26746  46106.66667  111479.3333  4116.666667  4693  166395.6667 
26782  46683  26972.4  0  0  73655.4 
26784  69160  21242  0  0  90402 
26785  69160  21933.6  0  0  91093.6 
26786  77805  21242  0  0  99047 
26788  74923.33333  21900.66667  0  0  96824 
26789  69160  23712  0  0  92872 
26791  64220  20042.28571  0  0  84262.28571 
26792  57633.33333  22065.33333  0  0  79698.66667 
26794  60515  23712  0  0  84227 
26795  57633.33333  22065.33333  0  0  79698.66667 
26797  69160  18772  0  0  87932 
26798  60515  23712  0  0  84227 
26800  51870  20254  0  0  72124 
26801  69160  21736  10585.71429  0  101481.7143 
26803  46106.66667  26099.66667  14408.33333  0  86614.66667 
27143  44460  57304  0  2964  104728 
27145  49400  44089.5  0  2964  96453.5 
27147  83362.5  18833.75  0  2593.5  104789.75 
27150  93860  32933.33333  4116.666667  3293.333333  134203.3333 
27152  83980  18722.6  4940  3705  111347.6 
27155  59280  41372.5  0  4075.5  104728 
27157  113620  79040  0  5681  198341 
27160  123500  69160  0  5928  198588 
27162  64220  55575  4940  3556.8  128291.8

60                  	Compendium  of  Project  Reports	 	
 	
27163  37544  21143.2  4940  2173.6  65800.8 
27165  60206.25  27633.125  3087.5  3581.5  94508.375 
27168  128440  44624.66667  0  4281.333333  177346 
27186  81510  23794.33333  4116.666667  2881.666667  112302.6667 
27189  91390  33345  0  3705  128440 
27191  106210  33098  0  3458  142766 
27267  30708.4186  15004.96279  7310.625581  14820  67844.00698 
27272  30381  15042.3  4841.2  14820  65084.5 
27275  31122  14754.13333  4841.2  14820  65537.33333 
27276  31122  15314  4841.2  14820  66097.2 
27278  31122  14754.13333  4841.2  14820  65537.33333 
27279  31122  14978.08  4841.2  14820  65761.28 
27280  26676  15153.45  4841.2  14820  61490.65 
27281  31122  10682.75  4841.2  14820  61465.95 
27282  31122  14754.13333  4841.2  14820  65537.33333 
27283  31122  15314  4841.2  14820  66097.2 
27284  31122  14754.13333  4841.2  14820  65537.33333 
27285  31122  14754.13333  4841.2  14820  65537.33333 
27288  31122  15832.7  4841.2  14820  66615.9 
27289  30486.85714  15003.48571  4841.2  14820  65151.54286 
27291  31122  15001.13333  4824.733333  14820  65767.86667 
27397  211333.2  20105.8  17092.4  22526.4  271057.8 
27398  174036.9841  15913.85714  33019.5873  19007.2381  241977.6667 
27399  185579.3333  15931.5  42628.08333  18969.6  263108.5167 
27400  205936.25  20089.33333  24339.79167  22559.33333  272924.7083 
27401  159879.5714  26517.21429  26810.08571  19004.88571  232211.7571 
27402  139752.6  15931.5  26799.5  18969.6  201453.2 
27403  232109.4286  17607.57143  17092.4  19336.57143  286145.9714 
27404  190313.5  20229.3  17092.4  22559.33333  250194.5333 
27405  168791.25  15917.25  26818.5  19000  230527 
27406  190719.2857  20095.21429  20642.14286  22524.04762  253980.6905 
27407  176111  20229.3  26815.96667  22526.4  245682.6667 
27408  175542.9  37173.5  26587.08  18969.6  258273.08 
27409  193154  123944.6  18772  22585.68  358456.28 
27410  226450.9722  20103.05556  17091.02778  22312.33333  285957.3889 
27657  49400  40755  43225  29640  163020 
27659  29640  18525  30875  24700  103740 
27660  26346.66667  29640  28816.66667  16466.66667  101270 
27661  187720  41990  29640  14820  274170 
27663  108680  33345  12350  14820  169195 
27666  59280  102505  49400  15437.5  226622.5 
27668  94848  0  9880  19760  124488 
27671  65866.66667  41825.33333  41166.66667  12350  161208.6667 
27743  235885  44608.2  2223  8299.2  291015.4 
27744  222300  44213  2223  8299.2  277035.2 
27745  216125  42422.25  2161.25  8336.25  269044.75 
27746  157462.5  43406.13333  2140.666667  8274.5  211283.8 
27747  214066.6667  44130.66667  2470  8274.5  268941.8333

61 	
Compendium  of  Project  Reports 	
27748  247000  44410.6  2223  8299.2  301932.8 
27749  182780  43136.08  2173.6  8269.56  236359.24 
27750  196828.125  42175.25  2161.25  8289.9375  249454.5625 
27751  222300  44410.6  2099.5  8521.5  277331.6 
27752  81664.375  42903.9  2161.25  8280.675  135010.2 
27753  223250  43206  2185  8265  276906 
27754  198835  44608.2  2223  8299.2  253965.4 
27755  175987.5  46139.6  2161.25  8336.25  232624.6 
27756  395200  0  2223  8892  406315 
27757  222300  42418.13333  2140.666667  8299.2  275158 	
 	
DATA PREPARATION 
â      	Missing  data and null values are treated	 	
â      	Study the correlation between the dependent variable (total cost) and independent             
variables(human labour, seed, fertilizer and other expenses).	 	
â     	For that we use heatmap.	 	
â     	From the heatmap it is clear that the dependent variable and independent variable 
are correlated.	 	
 	
 	
 
TOOLS AND LIBRARIES USED	 	
ï¶	 numpy  	
ï¶	 matplotlib.pyplot  	
ï¶	 pandas  	
ï¶	 seaborn  	
ï¶	 sklearn 	
  
RESULT WITH INFERENCE 
 	
â 	In this work, the impact of different factors that affect the cost of cultivation is 
considered and predicts how these factors affect the cost and forecast the cost.  	
â 	For that we use linear regression based prediction models. . Linear regression is a 
powerful tool for understanding and predicting the behavior of a variable. It is a

62                  	Compendium  of  Project  Reports	 	
 	
statistical approach for modeling the relationship between a dependent variable and 
a given set of independent variables.   	
â 	To check the accuracy we draw scatter diagram with actual value and predicted 
value.  	
â 	From the diagram it is clear that almost all the values lying in a straight line.  	
â 	It means that actual value and predicted values are almost equal.  	
â 	The model that we create is a perfect one.  	
 	
 	
â 	Calculating the regression coefficient and fit the data in the model 	 	
â 	We got the actual value.	 	
â 	predictedy = regr.predict([[34717.22222, 12034.38889, 28816.666670, 
5008.611111 ]])	 	
   	 	The predicted values corresponding to the above costs is ([80576.88889239])	 	
â  For this we t	ake 80% of  the data randomly for training the machine  and remaining 	
data for testing.	 	
â 	Using training data train the machine and construct a linear regression model .	 	
â 	Fit the test value in the model and we got the future predicted value.	 	
â 	In order  to find the accuracy of the model we calculate the error value	- mean 	
squared error.	 	
â 	We got mse=1.9770776563037933e	-05	 	
â 	It is evident that the model that we construct is correct and  using it we can predict 
the future value	 	
Actual Value and Predicted Value 
 	
Sl.No 	Y_test 	Y_pred 	
211 	127575.50000 	127575.50000053 	
180 	61914.66667 	61914.66666849 	
136 	45201.0000 	45200.99999913 	
91 	49940.31250 	49940.3124998 	
380 	93366.00000 	93366.00000076 	
158 	280839.00000 	280839.00000316 	
347 	75977.20000 	75977.20000032 	
56 	49605.83333 	49605.83332943 	
290 	63479.00000 	63479.00000111 	
145 	87573.85000 	87573.85000219

63 	
Compendium  of  Project  Reports 
 
The graphical representations are given below. 
 	
 	
 	
 	
 	
 	
 
 
CONCLUSION

64                  	Compendium  of  Project  Reports	 	
 
 
     	This paper was aimed to discuss the application of artificial intelligence in agriculture, 	
even  in  the  common  case  of  small  available  datasets.  The  approach  was  applied  and 
verified  in  the  specific  case  of  tapioca.  Also  aims  to  use  the  model  to  predict  other 
important crops. 
 
REFERENCE 	
1.   Report on Cost of Cultivation 2011-12 to 2020-21 published by Directorate of      Economics and 	
Statistics 	
2.  Multilinear regression model based studies

65 	
Compendium  of  Project  Reports 	
Utilising AI for accurate forecasting of 	
Keralaâs GSVA and NSVA	 	
A methodological Analysis and 	
Evaluation of benefits	 	
 	
Submitted  By 	
Smt. Praseeda Gopan, 	
Research Assistant

66                  	Compendium  of  Project  Reports	 	
 
Introduction 	
Forecasting  the  Gross  State  Value  Added  (GSVA)  and  Net  State  Value  Added 	
(NSVA)  of  Kerala  is  crucial  for  economic  planning  and  policymaking.  Accurate 
projections  of  these  figures  for  the  upcoming  financial  year  and  the  subsequent  year 
provide insights into the state's economic growth and development prospects. Traditionally, 
such  projections  are  made  based  on  the  analysis  of  historical  data,  economic  indicators, 
and  other  relevant  factors.  However,  with  the  advent  of  Artificial  Intelligence  (AI)  and  its 
advanced  analytical  capabilities,  there  is  a  growing  interest  in  exploring  its  potential  for 
more  accurate  and  efficient  projections.  In  this  report,  I  examine  how  AI  can  be  used  to 
project  the  GSVA/NSVA  of  Kerala  for  the  next  five  years  and  analyze  the  benefits  of 
using AI-based forecasting techniques over traditional methods. 	
 	
Objectives  	
The  primary  objective  of  this  report  is  to  utilize Artificial  Intelligence (AI)  techniques 	
for  the  projection  of  the  Gross  State  Domestic  Product  (GSDP)  and  Net  State  Domestic 
Product  (NSDP)  of  Kerala  for  the  next  five  financial  years.  Specifically,  I  aim  to  achieve 
the following: 	
1.	 	Develop  an  AI-based  forecasting  model  to  project  the  GSDP/NSDP  of  Kerala  for 
the upcoming financial years. 	
2.	 	Assess  the  impact  of  unforeseen  conditions  on  the  various  sectors  of  the  economy 
and evaluate the effect of these conditions on the projected GSDP/NSDP figures. 	
3.	 	Analyze  the  benefits  of  using  AI-based  forecasting  techniques  over  traditional 
methods for accurate and efficient projections of economic indicators. 	
4.	 	Provide  recommendations  for  policymakers  on  how  to  utilize  the  AI-based 
forecasting  model  to  make  informed  decisions  for  the  economic  development  of 
Kerala.  	
Methodology and method used 
The methodology employed in this report involved the following steps: 	
1.	 	Data  Collection:	  Historical  economic  data  of  Kerala  spanning  from  1970  to  2019 	
was collected from the Department.  	
2.	 	Data  Standardization:	  To  ensure  consistency  in  the  data,  all  economic  indicators 	
were brought to a common base year of 2011-12. 	
3.	 	Correlation  Analysis:	  To  identify  the  key  drivers  of  economic  growth,  heatmaps 	
were used to study the correlation between dependent and independent variables for 
both the primary sector and total GSVA.  	
4.	 	Model  Building:	  Linear  regression  models  were  built  to  calculate  the  regression 	
coefficient and fit the data for both the primary sector and total GSVA. The models 
were  evaluated  using  scatter  diagrams  to  compare  the  predicted  values  with  the 
actual values. 	
5.	 	Predictions:	  The  models  were  used  to  predict  the  GSVA/NSVA  figures  for  the 	
next  five  financial  years,  both  sector-wise  and  for  the  total  GSVA.  The  mean 
squared  error  was  calculated  for  both  models  using  Linear  Regression  to  evaluate 
the accuracy of the predictions.

67 	
Compendium  of  Project  Reports 
Datasets used 	
For  the  purpose  of  this  report,  I  have  used  historical  economic  data  of  Kerala 	
spanning  from  1970  to  2019.  It  is  important  to  note  that  the  base  year  for  the  economic 
data  has  been  shifted  several  times  during  this  period.  Therefore,  I  first  standardized  the 
data by  bringing  all  the  economic indicators  to  a  common  base  year  of 2011-12.  The  data 
was sourced from the Department of Economics and Statistics.  
 
Tools and libraries used 	
To  achieve  the  objective  of  projecting  the  Gross  State  Value  Added  (GSVA)  and 	
Net  State  Value  Added  (NSVA)  of  Kerala  using  Artificial  Intelligence  (AI),  I  have 
employed  a  variety  of  techniques.  In  particular,  I  utilized  machine  learning  algorithms, 
such as regression analysis and time-series forecasting, to build and test my models. I also 
used  data  visualization  techniques,  such  as  heatmaps,  to  study  the  correlation  between 
dependent  and  independent  variables.  By  examining  the  correlation  between  various 
economic indicators, I was able to identify the key drivers of economic growth in Kerala. 
Result with inference 
Firstly,  I  used  a  heatmap  to  study  the  correlation  between  the  dependent  variable 
(GSVA/NSVA)  and  independent  variables  for  both  the  primary  sector  and  total  GSVA. 
My  analysis  revealed  a  positive  correlation  between  the  dependent  variable  and 
independent variables in both sectors. 	
Primary Sector	 	
 	
 	
Total GSVA

68                  	Compendium  of  Project  Reports	 	
 	
Next, I built a linear regression model to calculate the regression coefficient and fit 	
the data for both the primary sector and total GSVA.  	
Primary Sector	 	
 	
Total GSVA 	
 	
To  evaluate  the  accuracy  of  my  predictions,  I  used  scatter  diagrams  to  compare  the 
predicted values with the actual values. My analysis showed that the plotted values were in 
a straight line, 	indicating	 that the model was a perfect fit. 	
Primary Sector

69 	
Compendium  of  Project  Reports 	
Total GSVA 	
 	
Finally,  I  conducted  sector-wise  predictions  as  well  as  predictions  for  the  total 	
GSVA,  and  calculated  the  mean  squared  error  for  both  models  using  Linear  Regression. 
My  analysis  revealed  that  there  was  no  significant  change  in  mean  squared  error  for  both 
models, indicating that the predictions were highly accurate. 	
Primary Sector	 	
 	
Total GSVA

70                  	Compendium  of  Project  Reports	 	
 
Conclusion 	
Overall,  the  methodology  allowed  me  to  identify  the  key  drivers  of  economic 	
growth  in  Kerala  and  build  an  AI-based  forecasting  model  that  accurately  projected  the 
GSVA/NSVA figures for the next five financial years. 
 
References 
 	
https://www.geeksforgeeks.org/linear-regression-python-implementation/ 
https://www.analyticsvidhya.com/blog/2022/02/linear-regression-with-python-implementation/

71 	
Compendium  of  Project  Reports 	
Analyzing and predicting the market 	
price of consumer products	
 	
Submitted  By 	
Sri. Nidhin Babu, 	
 Statistical Assistant Grade II

72                  	Compendium  of  Project  Reports	 	
 
Abstract 	
Using  machine  learning  algorithms,  this  study  proposes  a  methodology  for 
analysing  and  predicting  consumer  product  prices.  The  aim  of  the  study  is  to 
analyse the factors that influence the pricing of consumer products and to develop a 
predictive  model  that  can  be  used  to  estimate  the  prices  of  the  consumer  products. 
As part of the methodology, historical pricing data is collected from our department 
data library, data is cleaned and pre-processed, and machine learning algorithms are 
applied to build a predictive model. The study demonstrates the effectiveness of the 
proposed methodology by applying it to a case study of a consumer product. Based 
on the product's features and the market conditions, the predictive model is capable 
of accurately estimating the product's price. 
 	
Introduction 
 	
A  complex  task  like  analysing  and  predicting  the  price  of  consumer  products 	
involves various  factors such as consumer behavior, supply  and demand,  competition, and 
macroeconomic  influences.  In  order  to  analyse  and  predict  the  market  price  of  consumer 
products,  we  should  start  by  collecting  information  about  historical  prices,  sales  volumes, 
and market trends in order to establish a baseline. As a result of studying this data, we will 
be able to recognize patterns and correlations that can help us to understand the factors that 
influence prices in a more meaningful way.  	
Next,  the  opportunity  to  make  predictions  about  future  prices  using  statistical 	
models  and  machine  learning  algorithms  based  on  past  data  as  well  as  current  market 
conditions in order to make predictions about the future. A number of modeling techniques, 
such  as  linear  regression,  time-series  analysis,  and  neural  networks,  are  commonly  used. 
The accuracy of the predictions will depend greatly on the quality and relevance of the data 
as  well  as  the  complexity  of  the  market  dynamics  that  are trying  to  model  the quality  and 
relevance of the data we use as a starting point. Also, it is crucial to consider other factors 
that  may  have  an  impact  on  prices,  such  as  changes  in  regulations,  as  well  as  consumer 
preferences,  that  may  be  affecting  them.  In  general,  analysing  and  predicting  the  market 
price  of  consumer  products  is  a  challenging  but  important  process  for  businesses  and 
investors  who  are  interested  in  making  informed  decisions  about  pricing  and  strategic 
decisions. 	
 	
Grouping of consumer products  
 	
Depending  on  the  price  point  of  the  product,  consumers  can  be  categorized  into 	
various categories, including the following: 	
 
Economical  or  budgetary  products  are products  that  tend  to  have a  low  price  point 	
and cater to customers who are looking for products that are affordable. The basic features 
they have are often lacking and they may also compromise on the quality or design of their 
product. 	
A  mid-range  product  is  one  that  offers  a  great  balance  between  affordability  and 	
quality.  Usually,  these  products  are  more  expensive  than  economy  products  and  they  also 
offer more features and a better quality than economy products.

73 	
Compendium  of  Project  Reports 	
In  the  premium  category,  you  will  find  products  with  high-quality  features,  high-	
quality design, and advanced technology that are more expensive than mid-range products. 
In order to be competitive, they need to cater to customers who  are willing to spend more 
on luxury and exclusivity. 
 	
The  Luxury  Product  category  is  comprised  of  high-end  products  that  provide 	
premium features, superior quality, and exclusive designs to their users. These products are 
usually  offered  at  a  premium  price  and  are  geared  towards  customers  who  place  a  high 
value on luxury, status, and exclusivity	 	
Objective:- 	
âº	 Analyze the underlying patterns in consumer prices for various products. 	
âº	  to build  a system capable of finding patterns in data 	
âº	 A trend analysis examines the direction of the data. 	
âº	  Short term Forecasting of prices of consumer products 
 	
Literature review   	
A  similar  study  has  been  done  in   	user-inputted  text  descriptions  of  its  products, 	
including  details  like  product  category  name,  brand  name,  and  item  condition.  Using  this 
data, they have created a model that predicts the price of a product listed on Mercari[1].  	 
Another  paper[2]  addresses  the  problem  of  predicting  direction  of  movement  of 	
stock  and  stock  price  index  for  Indian  stock  markets.  The  study  compares  four  prediction 
models, Artificial Neural Network (ANN), Support Vector Machine (SVM), random forest 
and  naive-Bayes  with  two  approaches  for  input  to  these  models.  The  first  approach  for 
input data involves computation of ten technical parameters using stock trading data (open, 
high,  low  &  close  prices)  while  the  second  approach  focuses  on  representing  these 
technical parameters as trend deterministic data. Accuracy of each of the prediction models 
for each of the two input approaches is evaluated.  	
 
A study [3] that forecasts CPI which  is calculated in monthly periods each year to 	
anticipate  the  possibility  of a  spike  in  the  inflation  rate.  Forecasting  the  CPI  makes  use  of 
past  values,  commonly  known  as  time-series  data  (TSD).  One  method  to  assist  the 
forecasting  process  on  TSD  is  the  Autoregressive  Integrated  Moving  Average  (ARIMA). 
Another  method  that  can  also  be  used  for  forecasting  with  linear  data  problems  is  the 
Artificial  Neural  Network  (ANN).  This  study  compared  the  two  forecasting  methods 
between  ARIMA  and  ANN  by  predicting  the  Indonesian  CPI  value  from  January  - 
December  2018.  The  TSD  used  is  in  data  on  the  Indonesian  CPI  value  between  January 
2009  and  December  2017.  This  study  indicates  that  the  ANN  method  is  better  than 
ARIMA  because  it  produces  a  smaller  MSE  of  59.23.  However,  ARIMA  is  also  good 
because the two methods' forecast results are in the range of the CPI value. 	
 	
Methodology 
 	
The  objective  of  this  study  is  to  develop  a  machine  learning  model  for  predicting  the 	
short term forecast  of price of consumer products in Thiruvananthapuram  District, as well 
as to identify a machine learning technique that can be used to  classify items into various 
predefined categories.

74                  	Compendium  of  Project  Reports	 	
 	
 
 As part of the implementation process, the following steps will be taken: 	
â 	Collect  a  dataset  containing  time  series  data  containing  weekly  price  of  consumer 
products items as well as a dataset containing the names of items and their labels. 	
Table 1 	
 	
â 	Prepare the data by cleaning, formatting, and possibly removing irrelevant information. 	
â 	To begin with, I have selected an item for which I want to predict the price as a first step	. 	
Table 2 	
 	
Figure 1

75 	
Compendium  of  Project  Reports 	
â 	Understanding  the  underlying  pattern  of  time  series  data  by  decomposing  the 
data into various components of time series 	
   Figure 2 	
 	
Figure  2  clearly  shows  that  there  is  no  residual  (noise)  component  or  seasonality 	
component  in  the  data,  so  we  have  to  forecast  using  the  ARIMA  model.  ARIMA 
forecasting  requires  stationary  time  series  for  the  data  for  the  forecast.  The  Dickey  Fuller 
test  is  used  to  determine  whether  the  data  follows  the  condition  of  stationary.  With  null 
hypothesis as the data is stationary 	
1  .ADF Statistic:  -3.451925665842075 
2  P-Value:  0.009311069858978352 
3  n_lags  1 
4  n_lags 4.Num of observations used for ADF 	
Reggression and CriticalValues calculation 	
152 	
5  Critial Values:  1%, -3.474714913481481 
The null hypothesis has been at 1% level of Significance    	
It  is  possible  to  find  the  best  model  using  the  autoarima  function  present  in  the 
pmdarima  library.  The  auto  arima  function  calculates  the  Akakike  Information  Criteria 
(AIC)  for  an  ARIMA  model  based  on  the  different  parameters  that  are  present  in  the 
model.  It  is  decided  that  the  best  model  is  the  one  with  the  lowest  AIC  value.  It  is 
possible  to  find  the  best  model  using  the  autoarima  function  present  in  the  pmdarima 
library.  The  auto  arima  function  calculates  the  Akakike  Information  Criteria  (AIC)  for 
an  ARIMA  model  based  on  the different  parameters  that  are  present  in  the  model.  It  is 
decided that the best model is the one with the lowest AIC value.

76                  	Compendium  of  Project  Reports	 	
 	
âº	 Best model: ARIMA(1,0,1) intercept Total fit time: 10.640 seconds p=1,d=0,q=1 	
âº	 With minimum AIC value of 688.738688 	
To evaluate the model's performance, divide the data into training and test sets. 
âº	 70% (106 data points)data selected for Training the model 	
âº	 Remaining 30%(46 data points) for Testing  the model 	
âº	 Test RMSE: 1.909 	
Figure-3	 	
 	
Validity of the test 	
A  final  check  is  to  analyze  residual  errors  of  the  model.  Ideally,  Any  plot  for 	
residuals must be uncorrelated and the distribution of the residuals should follow a Gaussian 
distribution  with  a  zero  mean.  We  can  calculate  residuals  by  substracting  predicted  values 
from actuals as below. 
         Figure 4	 	
 	
         Figure 5

77 	
Compendium  of  Project  Reports 	
Using  the  model  once  it  has  been  validated,  it  can  be  used  to  predict  the  price  for 	
new extended period. It would be helpful if you could provide the model with the relevant 
features of the product and obtain a price prediction as a result.	 	
      In  order  to  make  an  accurate  prediction  out-of-sample,  it  is  important  to  keep  in 
mind  that  it  will  be  determined  by  the  quality  of  the  data,  the  relevance  of  the  selected 
features,  and  the  performance  of  the  machine  learning  algorithm  used  to  make  the 
prediction. Additionally,  the model should be continuously monitored and updated as new 
data becomes available, in order to ensure that the model remains  accurate and relevant in 
the future. 
Libraries and Tools used 	
There are a vast number of tools and libraries available for machine learning. Here are 	
some of the Libraries used for the study 
1.	 	Scikit-learn:	  A  popular  machine  learning  library  in  Python  for  tasks  such  as 	
classification, regression, and clustering. 	
2.	 	Pandas:	 A data manipulation library in Python that is often used to clean, transform, 	
and manipulate data for machine learning. 	
3.	 	NumPy:	  A  Python  library  for  numerical  computing  that  is  often  used  for  linear 	
algebra and array operations in machine learning. 	
4.	 	Matplotlib:	  A  Python  plotting  library  that  is  often  used  to  create  visualizations  of 	
data for machine learning. 	
5.	 	Seaborn:	 A data visualization library in Python that is built on top of Matplotlib and 	
provides a high-level interface for creating statistical graphics. 	
6.	 	pmdarima  :	is  a  Python  library  for  time  series  analysis  and  forecasting.  It  is  built  on 	
top of the statsmodels library and provides an interface for automating the process of 
selecting the best ARIMA model for a given time series.	 	
Results and Inference 
       	Figure 6 	
 	
A  comparison  has  been  made  between  the  date  available  after  out  of  sample 	
prediction and the date available after the prediction. the value displayed in the table 2.

78                  	Compendium  of  Project  Reports	 	
 
    Table 3 	
Date	 	Predicted value	 	Actual Value	 	
07-02-2023  19.8204  20 
14-02-2023  19.7029  20 
21-02-2023  19.5964  20 
28-02-2023  17.2371  16 
07-03-2023  16.7309  16 
14-03-2023  16.0825  15 
21-03-2023  15.7663  15 	
  Figure 7 	
 	
Conclusion 	
Although  the  performance  of  these  methods  depends  on  the  data  set  and  the 	
specific  features  chosen,  it  is  difficult  to  draw  definitive  conclusions  from  them.  The 
potential of machine learning techniques for predicting the Consumer products price under 
a variety of scenarios requires further research. 
 
Reference: 	
[1]   Kumar, A. (2020).Price Prediction using Machine Learning Regression â a case study 
[2]   Patel, J., Shah, S., Thakkar, P., & Kotecha, K. (2015). Predicting stock and stock price index 	
movement using trend deterministic data preparation and machine learning techniques. Expert 
Systems with Applications, 42(1), 259-268. 	
[3]  Time-series Modeling for Consumer Price Index Forecasting using Comparison Analysis of 	
AutoRegressive Integrated Moving Average and Artificial Neural Network Intan Yuniar 
Purbasari1 , Fetty Tri Anggraeny1 and Nindy Apsari Ardiningrum2 1 Department of Informatics, 
Faculty of Computer Science, Universitas Pembangunan Nasional âVeteranâ Jawa Timur, Raya 
Rungkut Madya, Surabaya, Indonesia

79 	
Compendium  of  Project  Reports 	
Crop yield prediction using Random 	
Forest Algorithm for different crops in 	
Kerala	
 	
Submitted  By 	
Smt. Neethymol Kurian	, Statistical Assistant Grade II

80                  	Compendium  of  Project  Reports	 	
 
Introduction	 	
  	
Crop  yield  prediction  is  a  critical  aspect  of  agriculture  that  can  help  farmers  make 	
informed  decisions  regarding  planting,  harvesting,  and  crop  management.  Accurate  crop 
yield  prediction  can  improve  crop  productivity  and  minimize  crop  losses,  which  can 
significantly impact a farmer's income and food security. 	
In  Kerala,  agriculture  is  an  essential  sector  of  the  economy,  contributing 	
significantly  to  the  state's  GDP.  The  state's  topography  and  climatic  conditions  make  it 
ideal  for  cultivating  a  wide  variety  of  crops  such  as  paddy,  coconut,  rubber,  spices,  and 
vegetables. However, due to unpredictable weather conditions and various pest and disease 
attacks, crop yield in Kerala is often volatile. 	
To address this issue, we propose using the Random Forest algorithm to predict the 	
crop  yield  of  different  crops  like  Coconut,  Tapioca,  Banana,                      Areca  nut,  Pepper, 
Ginger,  Turmeric  in  Kerala.  The  Random  Forest  algorithm  is  a  supervised  machine 
learning technique that has been shown to be effective in predicting yield in various crops. 
This  algorithm  can  handle  large datasets  with  many  variables  and  is  robust  to  outliers  and 
missing data. 	
 	
Objective 
 	
The  aim  of  this  project  is  to  develop  a  predictive  model  using  the  Random  Forest 	
algorithm  to  estimate  crop  yield  in  Kerala.  The  model  will  be  trained  on  crop  area  data, 
weather  data,  soil  data,  and  other  relevant  factors.  The  output  of  the  model  will  provide 
farmers  with  an  estimate  of  their  crop  yield  for  the  upcoming  year,  allowing them  to  plan 
and manage their crops more efficiently. 	
The  project's  outcomes  are  expected  to  help  farmers  in  Kerala  make  data-driven 	
decisions,  improve  crop  yield,  and  increase  their  income.  The  proposed  model  could  also 
be extended to other regions to enhance crop productivity and food security. 	
 	
Methodology  	
A.	 Dataset 
 
Data      plays      an  important  role  in    Machine    Learning.    To  design    and    perform  	
crop    yield    prediction    system    data    is  taken  from    various  districts  of    Kerala  state.  The 
description of the attributes mentioned in the dataset are mentioned in the table below. The 
data  used  in  the      project  is  taken  from  the  Agricultural  Statistics  2005-2020  Report 
published by DES, Kerala. (	Dataset  added  as  Annexure	)   	
Table  1:  Attributes  from  district  wise  crop  production  dataset 
 	
Sl.	 	
No.	 	Attribute  Description 	
 1. 	Agriculture Year	 	Five agriculture years data are taken starting from 
2015	-16 to 2019	-20.	 	
 2.	 	District	 	14 Districts of Kerala	 	
3 	Crop	 	Coconut, Tapioca, Banana, Areca nut, Pepper, Ginger, 
Turmeric   	 	
4. 	Area	 (Hectare)	 	Area considered for va	rious crops	 	
5. 	Production  (Tonnes)	 	Coconut 	
Production (Nuts) 	
Production of crops in the area specified.	 	
6. 	Cost of Cultivation	 (Rs/hectare)	 	Cost required for the cultivation of various crops per 
hectare 	
7.	 	Rainfall	 (mm)	 	Actual rainfall(mm) of all distri	cts in Kerala	 	
8.	 	Soil Composition	 	Nitrogen, Phosphorous and Potassium values

81 	
Compendium  of  Project  Reports 
 
B. Proposed System 	
A multiple linear regression model cannot be used in this case as the assumption of  	
linearity is not satisfied. 	
 	
 	 
Fig.1:  Checking  linear  relationship  between  yield  (dependant  variable)  and  Area,  Rainfall, 	
Cost  of  Cultivation(independent  variables  )	 	
There  are  many  non  linear  machine  learning  models  used  for  classification  and 	
regression.  Here  we  use  the  Random  Forest  approach  for  Crop  yield  prediction.    Random 
forest is a basically supervised learning algorithm. It creates decision trees on different data 
samples and then predict  the data from  each subset and then  by  voting   gives  better  the  
solution    for    the    system.  Instead  of  relying  on  one  decision  tree,  the random  forest  takes 
the prediction from each tree and based on the majority votes of predictions, it predicts the 
final output. The greater number of trees in the forest leads to higher accuracy and prevents 
the problem of over fitting.Given below is a pictorial representation of the model. 	
 	
Fig.2:	 Proposed  diagram  for  Crop  Yield      Prediction

82                  	Compendium  of  Project  Reports	 	
 	
Before applying the model to the data set apply the Data Preprocessing Techniques. 
 
ï· 	Detect  missing  values  in  the  dataset.  These  missing  values  are  replaced  with  the 
mode of soil composition (N) since it is an ordinal variable. 	
ï· 	Code the categorical variables â N, P, K. 	
ï· 	We have Area and Production in our data set . Calculate yield since our objective is 
to estimate the yield of various crops. 
 	
 	
 	
ï· 	Label Encoding is applied to Districts and Crops. 	
ï· 	The  dataset  will  be  split  to  two  datasets,  the  training  and  test  dataset.  The  data  is 
usually tend to split  unequally because training the model usually requires as much 
data points as possible. The common splits are 70/30 or 80/20 for train/test. 	
The  training  dataset  is  the  initial  dataset  used  to  train  ML  algorithm  to      learn  and 	
produce right predictions. (80% of dataset is training dataset). 	
The test  dataset, however, is used to assess  how well ML algorithm is trained with the 	
training  dataset.  You  canât  simply  reuse  the  training  dataset  in  the  testing  stage  because 
ML  algorithm  will  already  âknowâ  the  expected  output,  which  defeats  the  purpose  of 
testing the algorithm. (20% of dataset is testing dataset) 
Tools and Libraries used 	
 	
ï· 	Python Programming 	
ï· 	Libraries â numpy, pandas, matplotlib, sklearn, seaborn,  tkinter 
 	
Result and Analysis 
 	
 	
 	
ï· 	The  evaluation  metric  is  based  on   R^2  (coefficient  of  determination) regression 
score function, that will represent the proportion of the variance for items (crops) in

83 	
Compendium  of  Project  Reports 	
the  regression  model. R^2 score  shows  how  well  terms  (data  points)  fit  a  curve  or 
line.R^2 is  a  statistical  measure  between  0  and  1  which  calculates  how  similar  a 
regression  line  is  to  the  data  itâs  fitted  to.  If  itâs  a  1,  the  model  100%  predicts  the 
data variance; if itâs a 0, the model predicts none of the variance. 	
 
   	R2 score= 0.9114926858300401 	
 
Note	: 91% of the proportion of variance is explained by the Random Forest model for 	
Crop yield Prediction. 	
The  graph  below  shows  the  most  important  factors  that  affect  crop  yield.  The 	
importance  of  this  graph  is  that  we  can  eliminate  the  irrelevant  variables  based  on  its 
importance,  from the model and improve the accuracy of the model. 	
 	
                 	Fig.  3:  Feature  importance  of  variables  in  the  model 	
The  goal  of  the  project  is  to  build  a  graphical  user  interface  which  will  help  the 	
users,  policy  planners  to  predict  the  crop  yield  based  on  area,  cost  of  cultivation,  rainfall, 
soil composition	. 	
 	
 	
 	
Fig.4:  Graphical  User  Interface  that  shows  the  predicted  crop  yield  after  selection  of 	
different  parameters  by  user. 	
 	
As  shown    in    figure   4,    users    of   the application  can  see  the home  page  and  will 	
able  to  enter  the  details  such  as  the  district  code,  crop  code,  area,  cost,  rainfall,  soil

84                  	Compendium  of  Project  Reports	 	
 
composition    details  are  filled  on  the  home  page,  user  press  the  predict  button  then  the 
request will  send toward the server and the system gives a prediction using the model and 
trained under the random forest algorithm. The result  of the  prediction  of  the  crop  yield  
which  is  sent  to  the respective user and the unit of the crop yield is considered in tons.  
 
References 	
[1]    Kiran  Moraye,  Suyog  Nikam,  Smit  Thakkar,  Crop  Yield  Prediction  Using  Random  Forest 	
Algorithm  for  Major  Cities  in  Maharashtra  State,  International  Journal  of  Innovative 
Research  in  Computer  Science  &  Technology  (IJIRCST). 	
[2]   Sivanandhini  P,  Prakash  ,.Crop  Yield  Prediction  Analysis  using  Feed  Forward  and 	
Recurrent    Neural Network, International Journal of Innovative Research in Computer 
Science & Technology. 	
[3]   Crop Yield Prediction, https://www.kaggle.com/code/kushagranull/crop-yield-prediction 	
 
 
Sample dataset 
 	
Agric
ulture 
Year 	
District  Crop 	Area 	
_Hectare 	
Production
_Tonnes 	
Cost of 	
Cultivation_
Rs/Hectare	 	
Rainfall
_mm 	N  P  K 	
2015  Thiruvananthapuram	 	Tapioca  14585  536295  395217  2142.1  Low  Medium  Medium 	
2015  Thiruvananthapuram	 	Banana  2676  18338  489241  2142.1  Low  Medium  Medium 	
2015  Thiruvananthapuram	 Coconut  72340  633  371454  2142.1  Low  Medium  Medium 	
2015  Thiruvananthapuram	 	Arecanut	 	1036  643  263626  2142.1  Low  Medium  Medium 	
2015  Thiruvananthapuram	 	Pepper  2293  972  472349  2142.1  Low  Medium  Medium 	
2015  Thiruvananthapuram	 	Ginger  95  346  554412  2142.1  Low  Medium  Medium 	
2015  Thiruvananthapuram	 	Turmeric	 	73  185  487430  2142.1  Low  Medium  Medium 	
2016  Thiruvananthapuram	 	Tapioca  14628  520143  422946  981  Low  Medium  Medium 	
2016  Thiruvananthapuram	 	Banana  2776  19826  467942  981  Low  Medium  Medium

85 	
Compendium  of  Project  Reports 	
Changing trends in Cause of Deaths in 	
Kerala	 	
An analysis of cause of death data over 	
the years from 2012 to 2021 	 	
Submitted By 	
Kumari. Minu Merin Andrews, 	
Statistical Assistant Grade II

86                  	Compendium  of  Project  Reports	 	
 
 
Introduction 	
 	
The scheme of Medical Certification of Cause of Death (MCCD) was introduced in 	
the  country  under  the  provisions  of  Registration  of  Births  and  Deaths  (RBD)  Act,  1969. 
Medical  Practitioner  attending    the  deceased  at  the  time  of  death  fill  the  medical 
certification  form  recommended  by  World  Health  Organization.  The  forms  are  sent  to 
concerned  Registrar  of  Births  and  deaths  for    tabulation  .  The  cause  of  death  reported  are 
translated  in  to  medical  codes  contained  in  the  International  Classification  of  diseases 
published  by  the  World  Health  Organisation.  After  transmission  to  Additional  chief 
Registrars  office  the  State  subsequently  sends  data  to  the  Office  of  the    Registrar  General 
of India in the standardised format for National level consolidation. 
         	 	
In  Kerala  Medical  Certification  of  cause  of  death  scheme  has  been  introduced  in 	
Thiruvananthapuram,  Kollam,  Kochi  and  Kozhikode  corporations  and  Alappuzha 
Municipality. 
 
Objectives 
 	
Analysis  aimed  to  understand  how  causes  of    death  changed  over  years  from  2012 	
to  2021.  Analysis  has  been  made  on  the  incidence  of  mortality  due  to  nineteen  causes  of 
death disaggregated by gender and age.  	
 	
Objective was to understand 
  	
ï¶	 How the causes of death changed over time. 	
ï¶	 Biggest leading cause of death associated with each age group 	
ï¶	 	Break down of  deaths due to each cause in various age group	 	
 
Leading Causes of Death	 	
 	
Diseases  of  circulatory  system  accounted  for  highest  share  among  different  causes 	
of    deaths  from  year    2012  to  2020  .  There  has  been  a    change  in  this    trend  in  the  year 
2021  with  deaths  due  to  certain  infectious  and  parasitic  diseases  becoming  biggest  killer. 
Year  2021  observed  surging  share  of  deaths  due  to  infectious  and  parasitic  diseases 
accounting for 38.05% of total deaths among which viral diseases accounting for 94.16% . 
Diseases  of    circulatory  system  ranked  second  in  2021  with  20.69%  of  deaths.  Endocrine 
nutritional and metabolic diseases accounted for 11.57% of total deaths in  2021 and is the 
third  leading  cause  of  death.  Deaths  due  to  neoplasms  which  ranked  second    from  year 
2012  to  2019  and  third  in  2020  was  the fourth  leading  cause  of  death  in  2021  responsible 
for  9.65%  of  total  deaths.  Diseases  of  respiratory  system  was  the  fifth  leading  cause  in 
2021 with 4.83% of total deaths  in 2021. Diseases in the digestive system led to 4.12% of 
total  deaths  in  2021  and  is  the  sixth  leading  cause.  Injury,  Poisoning  and  certain  other 
external  causes  accounted  for  2.52%  and    diseases  of  genitourinary  system  accounted    
2.15% in 2021.

87 	
Compendium  of  Project  Reports 
 
Age Wise Analysis 
 	
ï¶	 From  2012  to  2020  leading  cause  of  death  in  age  group  70  and  above was  diseases  
of  circulatory  system.  In  2021  leading  cause  was  infectious  and  parasitic  diseases. 
43.47%  of  total  mortality  due  to  infectious  and  parasitic  diseases  was  in  people 
above 70  years of age. 44.63% of total deaths due to circulatory system occurred in  
this    age  group.  In  2021  42.47%  of  total    deaths  in  this  age  group  occurred  due  to 
infectious  and  parasitic  diseases,  23.7%  of  total  deaths  were  due  to  diseases  of 
circulatory system and 12.45% occurred due to endocrine, nutritional and metabolic 
diseases	.     	Percentage share of  deaths due to diseases of circulatory system in years 	
2012 and 2021 are 39.19% and 23.7% respectively. 	
ï¶	 In  65-69  age  group  also  leading  cause  of  death  in  2021  is  infectious  and  parasitic 
diseases. In this age group during 2021 percentage share of deaths due to infectious 
and  parasitic  diseases  is  38.34%.  14.78%  of  total  deaths  due  to  infectious  and 
parasitic  diseases  was  in  this  age  group  in  2021.  Diseases  of  circulatory  system 
which  was  the  leading  cause  of  death  in  this  age  group    from  2012  to  2020  is  the 
second  leading  cause  in  2021.    Diseases  of  circulatory  system  accounts  for  34.61% 
and  23.08%  of  total  deaths  in  this  age  group  for  years  2012  and  2021  respectively. 
Endocrine  nutritional  and  metabolic  diseases    is  the  third  leading  cause  in  2021 
whose percentage share is 13.81%. Neoplasms which was the second leading  cause 
from  2012  to  2019  and    third    leading  cause  in  2020  has  become  fourth  leading 
cause in 2021, with 9.91% of total deaths in this age group.	  	
ï¶	 Those  aged  between  55  and  64  years  diseases  of  circulatory  system  which  was  the 
leading cause from 2012 to 2020 was replaced by infectious and parasitic diseases in 
2021 accounting for  36.71% of total deaths in that age group.  Second leading cause 
is    diseases  of  circulatory  system  in  2021  with  share  of  21.08%.  Endocrine 
nutritional  and  metabolic  diseases  is  the  third  leading  cause  in  2021  whose share is 
13.1%.  Neoplasms  which  was  the  second  leading  cause  from  2012  to  2019  and  
third    leading  cause  in  2020  has  become  fourth  leading  cause  in  2021  with  12.36% 
share.  	
ï¶	  Certain infectious and parasitic diseases which was the fifth leading cause of death 
in  45-54  age  group  with  a  share  of  8.94%  in  2012    has  become  leading  cause  of 
death  in  2021  accounting  for  36.42%  deaths.  Neoplasms  was  the  second  leading 	
 	
 	
 	
 	
According  to  data  from  Medical 
Certification  of  Cause  of  death  viral 
diseas	es  accounted  for  35.83%  of  	
total  deaths  in  year  2021.  35.46% 
and  36.42%  of  total  male  and  total 
female deaths respectively.

88                  	Compendium  of  Project  Reports	 	
 	
cause  of  death  in  years  2012  to  2016  and  2019  to  2020.  In  2017  and  2018 
Neoplasms  was  the  leading  cause  of  death  in  this  age  group.  Share  of  Neoplasm  is 
19.51%,  22.19%,21.86%,16.75%  and  11.76%  for  years  2012,2018,2019,2020  and 
2021 respectively. Diseases of  digestive system was the third leading cause of death 
from  years  2012  to  2017  and  fourth  leading  cause  in  years  2019  and  2020.  During 
2021,  Diseases  of  circulatory  system,  Neoplasms,  Endocrine  nutritional  and 
metabolic  diseases,  Diseases  of  digestive  system  ranked  second,  third,  fourth  and 
fifth respectively. 	
ï¶	 For  those  aged  between  35-44    leading  causes  of  deaths  were  neoplasms  and 
diseases  of  circulatory  system.  In  2016  leading  cause  was  diseases  of  digestive 
system.  Diseases  of  digestive  system  ranked  third  for  all  other  years  from  2012    to  
2020.  During  2021  Certain  infectious  and  parasitic  diseases,  diseases  of  circulatory 
systems,  neoplasms  and    diseases  of  digestive  system  ranked  first,  second,  third, 
fourth and fifth respectively. 	
ï¶	 In  age  group  25-34  leading  cause  of  death  for  years  2012,2013,2015  to  2020  and 
second  leading  cause  of  death  in  2014  was  injury  poisoning  and  certain  other 
external  causes.  In  2012  percentage  share  of  this  cause  was  18.82  %  and  in  2021 
share  is  16.74%,  20.26%  of  total  male  deaths  and  11.83%  of  total  female  deaths. 	
Neoplasms  was  the  second  leading 
cause  in  2013  and  2015  to  2020. 
Share  of  neoplasm  in  years  2012, 
2021  is  14.8%  and  16.74% 
respectively. Diseases of circulatory 	
system ranked second in 2012 and third from 2013 to 2017 and fourth from 2018 to 
2021.  In  2012  and  2021  percentage  share  due  to  diseases  of  circulatory  system  in 
this  age  group  was  16.76%  and  8.37%  respectively	. 	In  2021  leading  cause  was 	
certain  infectious  and  parasitic  diseases  accounting  for  30.34%  of    total  deaths  in 
that age group. 	
ï¶	 In  age  group  15-24  leading  cause  of  death  is  injury,  poisoning  and  certain  other 
external causes for all years from 2012 to 2021. During  year 2021 injury, poisoning 
and  certain  other  external  causes  accounted  for  21.01%  of  total  deaths  in  this  age 
group,  29.91%  of  total  male  deaths  and  10.53%  of  total  female  deaths  in  this  age 
group  occurred  due  to  injury,  poisoning  and  certain  other  external  causes.  There  is 
significant  difference  between  male  and  female  proportion  of  deaths  due  to  injury, 
poisoning  and  certain  other  external  causes.  In  2021  second  leading  cause  of  death 
is  neoplasms  (19.32%  of  total  deaths)  and  third  leading  cause  is  infectious  and 
parasitic diseases (17.87% of  total deaths). 	
ï¶	 In  5-14  age  group  leading  cause  of  death  is  neoplasms  for  all  years  from  2012  to 
2021.  During  2021  percentage  share  of  neoplasms  to  total  deaths  was  23.22%. 
Leukaemia  accounts  for  12.8%  of  total  deaths  in  5-14  age  group  (19.05%  of  total 
male  deaths  and  6.6%  of  total  female  deaths).  In  2021  second  leading  cause  is 
certain  infectious  and  parasitic diseases  (16.11%  deaths  )  and  third  leading  cause  is 
injury, poisoning and certain other external causes (13.27% deaths). 
 
In  the  year  2021  Certain  infectious  and 
parasitic  diseases  was  the    leading  cause  of 
death  for  all  above  25  years  age.

89 	
Compendium  of  Project  Reports 	
ï¶	 For  children  of  age  group  1-4    Congenital  Malformations,  Deformations  and 
Chromosomal Abnormalities was the leading cause from year 2015 to 2021. Certain 
Infectious and parasitic diseases was the third leading cause during year 2021. 	
ï¶	 For children below 1  year certain conditions occurring in perinatal period is leading 
cause  and  second  leading  cause  was  congenital  malformations,  deformations  and 
chromosomal  abnormalities  for  years  from  2012  to  2021.  Certain  Infectious  and 
parasitic  diseases  has  become  the  third  leading  cause  of  death  in  2021  in  this  age 
group. 
 	
CERTAIN INFECTIOUS AND PARASITIC DISEASES	 	
 	
 	
 	
Proportion  of  deaths    due  to  certain  infectious  and  parasitic  diseases  which 	
accounted  for  6.93%    of    total    deaths  in  2012  has    increased  to  10.73%  in  2020    and   
38.05%  in  2021.  Intestinal  infectious  diseases,  Tuberculosis,  Other  bacterial  diseases, 
Infections  with  a  predominantly  sexual  mode  of  transmission  ,  Viral  diseases,  Protozoal 
diseases,  Other    certain  infectious  &  parasitic  diseases  and  late  effects  of  infectious  & 
parasitic  diseases    are    the    causes    coming  under  this  group.  In  2020  and    2021  viral 
diseases  accounted  for  8.16%  and  35.83%    of  total  deaths  .  Not  much  gender  gap  is 
observed  in  the  percentage  share  of    deaths  to  total  deaths  in  2021,  37.56%  of  total  male 
deaths and 38.84% of  total female deaths . Percentage  share of deaths increased  in all age 
groups  in  year 2021 compared  to 2020 and has become the leading cause of death for all 
those  above  25  years  of  age.  Out  of  total    deaths  due  to  infectious  and  parasitic  diseases 
majority  share  is  for  people above  70  years,  43.47%.  14.78%  in  65-69  age  group,  21.24% 
in 55-64 age group, 10.83% in 45-54 age group and 4.47 % in 35-44 years of age.

90                  	Compendium  of  Project  Reports	 	
 	
 	 
 
Endocrine, Nutritional and Metabolic diseases

91 	
Compendium  of  Project  Reports 	
Percentage  share  of    deaths  due  to  Endocrine,  Nutritional  and  Metabolic  diseases 	
has been steadily increasing from year 2016 to 2020. In 2021 share decreased from 15.58% 
to 11.57%.  
 	
 	
 	
Percentage of deaths due to Endocrine, Nutritional and Metabolic  diseases to total 	
deaths  has been  increasing for both  gender from  2016 to 2020. From  year 2016 to 2020 
there  has  been  an  increase  from  10.4%    to  18%  (increase  of  7.6  percentage  points)  for 
female    and    an  increase  from  8.2%    to  14.13%  (increase  of  5.93  percentage  points)  for 
male  .  During  year  2021  Diabetes  Mellitus    accounts  for  94.23%  of  total  deaths  due  to 
endocrine, nutritional and metabolic diseases . Share of  diabetes Mellitus to total deaths in 
the year 2021 is 10.9%.

92                  	Compendium  of  Project  Reports	 	
 
Neoplasm 
 	
 	
 	
Percentage  of  deaths  due  to    Neoplasms  to  total    deaths  in  15-24    age    group  	
increased in 2021 compared to 2020. Percentage share of deaths due to neoplasms in 15-24 
age group was 14.68% and 19.32% for  years 2020 and 2021 respectively.  Neoplasms was 
the  second  leading  cause  of  death  in  age  group  25-34  for  years  2013  and  years  2015  to 
2020. In 2021 neoplasms accounted for 16.74% of total deaths in age group 25-34 (12.05% 
of  total male deaths and 23.3% of total female deaths).	 	
 	
 	
 	
During  the  year  2021  among  the  âNeoplasmâ  deaths,  âMalignant  Neoplasm  of 	
Digestive  Organsâ  accounts  for  the  highest  mortality  (29.71%),  followed  by  âMalignant 
Neoplasm  of  Respiratory  and  Intrathoracic  Organsâ  (17.65%),âMalignant  Neoplasms  of 
Lymphoid, Haematopoietic& other related tissueâ (17.45%),âMalignant  neoplasms of bone, 
mesothelial    and  soft  tissue,  skin  and  breastâ  (10.42  %),    âMalignant  neoplasms  of 
genitourinary  organsâ  (9.73%)  and  âMalignant  neoplasms  of  lip,  oral  cavity  and  pharynxâ

93 	
Compendium  of  Project  Reports 
(5.79  %)  and  âMalignant  neoplasms  of  other    unspecified  sitesâ  (5.47  %)  are  other  major 
causes. 	
In  2021,  27.18%  of    deaths  due  to  Neoplasms  occurred  in  those  above  70  years  of 	
age  ,  15.06  %  in  65-69  years  age  group,  28.19%  in  55-64  years  age  group  and  13.79%  in 
45-54,  6.97  %  in  35-44  years,  3.22%  in  25-34  age  group      ,  2.3%  in  15-24  age  group,       
1.41% in 5-14 age group. 
 	
 	
 	 	
 
Methodology  Used 
 
Data validation. 
Data processing for finding proportions and percentage share in each age group 
Exploratory and descriptive data aanalysis 
 
Tools  and Libraries  used  
Project was done in R Markdown. 
 	
ï¶	 Conditional statements and loops were used for validating data. 	
ï¶	 Functions in readxl, openxlsx, dplyr, tidyr packages were used for importing, 
processing and exporting data . 	
ï¶	 Exploratory data analysis was done using ggplot2 package. 	
 
Conclusion 
 	
In  the  year  2021  infectious  and  parasitic  diseases  has  become  the  leading  cause  of 	
death  in  Kerala  for  people  aged  above  25  years.  This  shows  an  increasing  share  of 
communicable diseases like viral diseases to total deaths.

94                  	Compendium  of  Project  Reports

95 	
Compendium  of  Project  Reports 	
Time series Analysis of Gross Value 	
Added of Organized Manufacturing 	
Sector of India and Kerala 	
 	
Submitted By 	
Kumari. Baby Sindhu, 	
Statistical Assistant Grade II

96                  	Compendium  of  Project  Reports	 	
 
 	
Abstract 	
 	
This  project  aims  to  provide  an  analysis  on  the  time  series  data  of    Gross  Value 
Added  (GVA)  of  Organized  Manufacturing  Sector  of  India  and  Kerala  using  the 
estimated  data  of  Annual  Survey  of  Industries  conducted  by  Ministry  of  Statistics 
and  Programme  Implementation  from  1980-81  to  2019-20.  The  Box-Jenkins 
procedure  is  used  to  determine  the  ARIMA  (Auto  Regressive  Integrated  Moving 
Average  )  model  of  the  time  series  data.    The  model  is  used  to  forecast  the 
performance of the GVA in the upcoming years. 	
 	
1.	 	Introduction 	
 	
Industries plays a pertinent role in the economic growth of the nation.  In particular, 	
the contribution of Organised Manufacturing sector are imperative in the countryâs overall, 
and  especially  its  economic  development.  The  expansion  of  a  countryâs  manufacturing 
industries  is  used  to  gauge  its  economic  strength.    Gross  Value  Added  (GVA)  is  an 
economic productivity metric that measures the contribution of the manufacturing sector to 
the  economy.    It  is  the  difference  of  the  Gross  Output  and  Gross  Input  of  the 
manufacturing  industries	. Analyzing  time  series  data  and  Forecasting  economic  model  is 	
essential  for  the  policy  makers  in  developing  new  strategies  for  further  development  of 
economy.  	
  	
2.	 	Objective 
 
The  objective  of  this  project  work  was  to  develop  forecasting  models  for  predicting 	
the Gross Value Added (GVA) of the Organized Manufacturing Sector of India and Kerala 
from  the  historical  data.    Box-Jenkins  methodology  has  been  followed  to  model  the GVA 
through time series with the ARIMA univariate approach.  The purpose of this time series 
analysis  is  to  extract  information  about  the  time  series  observations  and  use  these 
information  for  decision  making  purpose.    The  dataset  used  in  this  study  comprises  of 
GVA of organized manufacturing sector in India and Kerala from 1980-81 to 2019-20 (40 
years) and the entire historical data is used to build up a regression model and forecast the 
future.    This  provides  the  consequences  and  insights  of  features  of  the  given  dataset  that 
changes over time.  The analysis has been done using Python Programming. 	  	
3.	 	Methodology of Time series analysis 	
 	
Time  series  analysis  consists  of  various  techniques  which  is  used  to  reproduce  the 	
observations  with  a  mathematical  model  and  hence  predicts  the  estimate  of  the  variable 
using  the  relevant  historical  data.  In  this  study  Box  Jenkins  Auto  Regressive  Integrated 
Moving Average (ARIMA) method is adopted for building model and forecasting. A usual 
time series data comprises of the following components which are the factors affecting the 
values of the phenomenon:  	
a.	 	Secular  Trend  or  Long-Term  Movement:   	By  trend  we  mean  the  general 	
tendency of the data to increase or decrease during a long period of time. 	
b.	 	Seasonal  variation:	  These  variations  in  a  time  series  are  due  to  forces  which 	
operate in a regular and periodic manner over a span of less than a year.

97 	
Compendium  of  Project  Reports 	
c.	 	Cycli	c:  The  oscillatory  movements  in  time  series  with  period  of  oscillation 	
more than one year.  	
d.	 	Irregular  component	:  Apart  from  the  regular  variation  a  series  may  contain 	
random  or  irregular  factors  which  are  not  counted  by  secular  trends,  seasonal 
and cyclic variations. 
 	
The  ARIMA  models  which  is  built  in  this  study  assumes  that  the  new  observation 	
depend purely on the weighted linear combination of past values. It assumes that the series 
is stationary i.e., mean, variance and autocorrelation are constant over time.  If the series is 
not  stationary  differencing  method  is  adopted  to  transform  the  series  in  to  stationary.  The 
following are the models:  	
 	
Autoregressive (AR) Model 
 
            	In  an  autoregressive  mode    we  forecast  the  variable  of  interest  using  a  linear 	
combination  of  past  values  of  the  variable.  The  term  autoregressive  indicates  that  it  is  a 
regression of the variable against itself:  	
 	 	
where 	Ýt is white noise. This is like a multiple regression but with lagged values of 	
yt as predictors. This as an AR(	p) model, an autoregressive model of order 	p. 	
 
Moving Average (MA) Model	 	
 
           Rather than using past values of the forecast variable in a regression, a moving 
average model uses past forecast errors in a regression-like model: 
 	
 	 
 This as an MA(	q) model, a moving average model of order 	q (fpp2). 	
 
Autoregressive Integrated Moving Average (ARIMA) Model	 	
 
           If we combine differencing with auto regression and a moving average model, we 
obtain a non seasonal ARIMA model: 
 	
 	
Where 	UPâ²  is  the  differenced  series.  This  ARIMA  (	L, @, M)  model,  where 	L  is  the 	
number  of  autoregressive  terms, 	@  is  the  degree  of  differencing  and 	M  is  the  number  of 	
moving average terms.	 	
3.1 Time series Analysis process 
 
        Using  Box-Jenkins  methodology  following  procedures  were  followed  for 
identification of model, Parameter estimation/calibration and Model testing/validation.

98                  	Compendium  of  Project  Reports	 	
 
The basic steps followed is depicted in the flow chart below: 	
 	
The following procedures is adopted for the analysis: 
i.     P	lotting time series data	: Time series data was plotted to identify the nature and the 	
trend of the variable.    	
ii.   Stationarity of time series	: A time series is said to be stationary if mean, variance and 	
autocorrelation remains constant over time. Graphically, the time series is stationary if 
the  correlogram/ACF  (Auto  Correlation  Function)  (a  plot  of  autocorrelation 
coefficients on the vertical axis with different lags on horizontal axis) dies down fairly 
quickly and the series is non-stationary if decay is very slow.  	
 
 Stationarity can also be checked using Augmented Dickey-Fuller (ADF) test where, 
 
H0:  	The  null  hypothesis:	 It  is  a  statement  about  the  population  that  either  is 	
believed  to  be  true  or  is  used  to  put  forth  an  argument  unless  it  can  be 
shown to be incorrect beyond a reasonable doubt. 	
H1:  	The  alternative  hypothesis:	 It  is  a  claim  about  the  population  that  is  contradictory 	
to H	0 and what we conclude when we reject H	0. 	
#Ho: It is non-stationary 
#H1: It is stationary 
  
If p value >0.05, Fail to reject H0 
If p value <0.05, Accept H1 
 
Methods for transforming non-stationary data to stationary data  
 	
The major methods used for  converting non-stationary to stationary for time series 	
modelling are:

99 	
Compendium  of  Project  Reports 	
ï· 	Detrending 	
ï· 	Differencing 	
ï· 	Transformation  	
 
In  this  study  differencing  method  is  used  for  transforming  data  to  stationary 
data.  
 
iii.	 	Model Identification 	
 
   The  model  identification  is  done  using  Auto  Correlation  Function  (ACF)  and 
Partial Auto Correlation Function (PACF).  
 
 Auto Correlation Function (ACF)/Correlogram 
   ACF indicates the memory of the stochastic process.  It indicates the dependence of 
the  value  of  the  variable  on  the previous  value.    It  is 	a  plot  of  autocorrelation  coefficients 	
on the vertical axis with different lags on horizontal axis. 
 
 Partial Correlation Coefficient Function (PACF) 
  	PACF  indicates  the  dependence  of  value  of  the  variable  on  the  previous  values 	
when the dependence on all other variables are removed. 
 
Auto Regressive Model 
AR(p) models can be identified as follows: 	
i. 	Exponential or sinusoidal decay in ACF. 	
ii.	 	PACF cuts off after a few say, âpâ lags. 	
 Moving Average Model 
 MA(q) models can be identified as follows: 	
i. 	Decaying of PACF (either exponential or in a dampened sine wave) 	
ii.	 	ACF cuts off after a few say, âqâ lags. 	
 Auto Regressive Integrated Moving Average Model 
  	If both ACF and PACF decay gradually ARIMA model is used. In ARIMA (p,d,q) 	
parameters p, q and d are non-negative integers, p is the order (number of time lags) of the 
autoregressive  model,  d  is  the  degree  of  differencing  (number  of  times  the  data  have  had 
past values subtracted), and q is the order of the moving average model.   
 
iv.     Model estimation 
  	Akaike  Information  criterion  (AIC)  and  Bayesian  Information  Criterion  (BIC)  are 	
used  to  test  how  well  the  model  fit  the  data.    The  best  model  is  identified  based  on  high 
value  of  AIC  and  low  value  of  BIC.      The  AIC  and  BIC  is  described  mathematically  as 
follows:

100                  	Compendium  of  Project  Reports	 	
 
v. Model Validation 
  	Residual Analysis is done to evaluate the accuracy of the model.  The accuracy of 	
the model can be assessed by measuring the difference between predicted value and actual 
value.  In ARIMA the above difference is the error term and it behave as white noise, i.e, 
zero mean, a constant variance and no correlation.  The error term should also be normally 
distributed.     
 
 Box-Ljung test 
  	Box-Ljung test is a diagnostic tool applied to residuals to test the lack of fit of time 	
series.  It examines the autocorrelations of the residuals.  The null hypothesis is that series 
is uncorrelated.  If p value >0.05 then the residuals are uncorrelated.   
 
 Shapiro âWilk test for Normality check 
    The  Shapiro-Wilk  test  for  normality  is  a  diagnostic  tool  of  statistics  to  assess  the 
normality of  the  data.    The  null  hypothesis  is  taken  as  H0:  The  residuals  are  normally 
distributed.  If p value >0.05 then normality assumption is satisfied.   
 
vi. Forecasting 
       After  the  validation  procedure  the  model  can  be  used  for  forecasting  the  future 
values.  
vii	. Measure of Accuracy	  	
   To measure the accuracy of the forecasted data 	Mean Absolute Percentage Error 	
(MAPE) 	is used.   	
 	
3.2 Time series Analysis  
   The  time  series  analysis  is  conducted  on  the  data  comprising  of  the  Gross  Value 
Added (GVA) of industries of India and Kerala in the organised manufacturing sector from 
1980-81 to 2019-20.  The GVA is expressed in Rupee lakhs.  The data has been procured 
from the official site of Ministry of Statistics and Programme Implementation (MOSPI).

101 	
Compendium  of  Project  Reports 
3.2.1	 	Time Series Analysis for predicting the GVA of Organised manufacturing 
sector of India 	
1.	  Plotting Time series Data 	
 The trend of time series data of GVA of India in 40 years is shown in 	Figure  1. 	
 	
         Figure  1 
 	
2.	 Decomposition of time series data  	
 	
 	
Figure  2

102                  	Compendium  of  Project  Reports	 	
 	
 	
Figure  3 	
 3. Training and Test Data  
 
         The  whole  data  set  is  split  into  training  and  test  data.    The  data  from  1980-81  to 
2015-16 is selected as train data and 2016-17 to 2019-20 as test data.  The whole ARIMA 
methodology is performed in the train data to build a model and further the model is tested 
in the test data and after the validation and accuracy checking the model is used to forecast 
the future data.	 	
 	
 	
Figure  4

103 	
Compendium  of  Project  Reports 
    4. Stationarity Check	 	
                 Auto Correlation function  	
 	
 	
Figure  5 	
 	
Partial Autocorrelation Function 	
 	
Figure  6 	
Dickey-Fuller Test

104                  	Compendium  of  Project  Reports	 	
 	
 	
 	
Figure  7 	
 	
       The ACF, PACF curve, Rolling mean and Standard deviation curve and Dickey-Fuller 
Test  shows  that  the  data  is  non-Stationary.    Further  differencing  method  was  adopted  to 
transform the data in to stationary data. 
 
Differencing for transforming the data into stationary data. 
       The  data  was  undergone  differencing  twice  to  transform  to  stationary  data.    The 
Dickey- Fuller test result after differencing is as follows: 
 	
 	
The results show that p value <0.05.  So the data is stationary. 	
Distribution of Data after Second Differencing 	
 	
 	
Figure  8

105 	
Compendium  of  Project  Reports 	
Rolling Mean & Standard Deviation 	
 	
                          Figure  9 	
 
 
5. Identifying the model using ACF and PACF  
As per the ACF and PACF curve the model identified was ARIMA (1, 2, 1) 
 	
 	
 	
Figure  10 	
 	
 	
Figure  11

106                  	Compendium  of  Project  Reports	 	
 	
 
 	
6. Plotting actual and predicted value of trained data 
 
 	
 	
Figure  12 	
 	
 	
 
7.Model Validation using Box-Ljung Test 
 
The test shows that p value >0.05 and hence the residuals are uncorrelated.  
 
8. In sample forecast of GVA 	
 	
Year  Actual GVA 	
(in Rupee lakhs) 	
Predicted GVA (in 	
Rupee Lakhs) 	
2016-17  136805409  137428200 
2017-18  146697043  147741200 
2018-19  153801928  157994500 
2019-20  148574512  168264500 	
 
9. Residual Analysis 
 
Shapiro âWilk Test for Normality 
 
The test shows that p value >0.05 therefore the residuals are normally distributed.

107 	
Compendium  of  Project  Reports 	
 	
Figure  13 	
 	
10. Measure of Accuracy 
 
         To  measure  the  accuracy  of  the  forecasted  data 	Mean  Absolute  Percentage  Error 	
(MAPE) 	is  used.    The  MAPE  value  is  as  follows  which  shows  4%  error  in  predicted  and 	
actual data.  
 	
 	
11. Out Sample Forecasting 
 	
Year  Predicted GVA (in Rupee 	
Lakhs) 	
2020-21  178529800 
2021-22  188796500 
2022-23  199062800 
2023-24  209329200 	
 
3.2.2 Time Series Analysis for predicting the GVA of Organised manufacturing 
sector of Kerala 
1. Plotting the GVA of Kerala

108                  	Compendium  of  Project  Reports	 	
 	
Figure  14 	
2. Decomposition of time series data  	
 	
Figure  15 	
 	
Figure  16 	
3. Trained Data and Test Data 
          The  data  from  1980-81  to  2015-16  is  selected  as  train  data  and  2016-17  to  2019-20 
as test data.   	
 	
Figure  17

109 	
Compendium  of  Project  Reports 
12. Stationarity check 
Auto Correlation function 	
 	
Figure  18 	
Partial Autocorrelation Function 	
 	
Figure  19 	
Dickey-Fuller Test 	
 	
 	
 	
Figure  20

110                  	Compendium  of  Project  Reports	 	
 
       The ACF, PACF curve, Rolling mean and Standard deviation curve and Dickey-Fuller 
Test  shows  that  the  data  is  non-Stationary.    Further  differencing  method  was  adopted  to 
transform the data in to stationary data. 
 
Differencing for transforming the data into stationary data. 
 
       The  data  was  undergone  differencing  thrice  to  transform  to  stationary  data.    The 
Dickey- Fuller test result after differencing is as follows: 
 	
 	
   The results show that p value <0.05.  So the data is stationary. 
Distribution of Data after Third Differencing 	
 	
Figure  21 	
Rolling Mean & Standard Deviation 	
 	
             Figure 22

111 	
Compendium  of  Project  Reports 
13. Identifying the model using ACF and PACF 
As per the ACF and PACF curve the model identified was ARIMA (4, 3, 1) 	
 	
 	
Figure 23 	
 	
 	
Figure 24 	
 	
14. Plotting actual and predicted value of trained data 
 	
 	
Figure 25

112                  	Compendium  of  Project  Reports	 	
 	
 	
 
15. Model Validation using Box-Ljung Test 
 
The test shows that p value >0.05 and hence the residuals are uncorrelated.  	
 	
16. In sample forecast of GVA 
 	
Year  Actual GVA 	
(in Rupee lakhs) 	
Predicted GVA (in 	
Rupee Lakhs) 	
2016-17  2236188  2004238 
2017-18  2183551  2146227 
2018-19  2337876  2444895 
2019-20  2252678  2479974 	
 
17. Residual Analysis  
Shapiro âWilk Test for Normality 
 
       The test shows that p value >0.05 therefore the residuals are normally distributed.

113 	
Compendium  of  Project  Reports 	
 	
 	
Figure 26 	 
18. Measure of Accuracy 
 
          To  measure  the  accuracy  of  the  forecasted  data 	Mean  Absolute  Percentage  Error 	
(MAPE) 	is used.  The MAPE value is as follows which shows 6.8% error in predicted and 	
actual data.  	
 	
19. Out Sample Forecasting 
 	
Year  Predicted GVA (in Rupee 	
Lakhs) 	
2020-21  2933417 
2021-22  3160725 
2022-23  3267800 
2023-24  3718621 	
 
4. 	Dataset used  	 	
Gross Value Added (GVA) of Organized Manufacturing sector of Kerala from 	
1980-81 to 2019-20 	
 	
Financial Year  Gross Value Added (GVA) of 	
Kerala (in Rupee lakhs ) 	
1980-1981  44453 
1981-1982  49663 
1982-1983  54563 
1983-1984  59976 
1984-1985  77272 
1985-1986  77841

114                  	Compendium  of  Project  Reports	 	
 	
1986-1987  82298 
1987-1988  102975 
1988-1989  111358 
1989-1990  160235 
1990-1991  138461 
1991-1992  190971 
1992-1993  207178 
1993-1994  203119 
1994-1995  254675 
1995-1996  344068 
1996-1997  364840 
1997-1998  376065 
1998-1999  463724 
1999-2000  414630 
2000-2001  418162 
2001-2002  403543 
2002-2003  429000 
2003-2004  477401 
2004-2005  482950 
2005-2006  523307 
2006-2007  436893 
2007-2008  682240 
2008-2009  882107 
2009-2010  828506 
2010-2011  1004010 
2011-2012  1079692 
2012-2013  1335966 
2013-2014  1526293 
2014-2015  1421515 
2015-2016  1829940 
2016-2017  2236188 
2017-2018  2183551 
2018-2019  2337876 
2019-2020  2252678 	
              Source: www.mospi.gov.in 	 	
Gross Value Added (GVA) of Organized Manufacturing sector of India from         	
1980-81 to 2019-20 	
 	
Year  Gross Value Added (GVA) in Rupee lakhs) 	
1980-1981  1407076 
1981-1982  1668146 
1982-1983  1914118 
1983-1984  2352049 
1984-1985  2494163 
1985-1986  2696866 
1986-1987  3019912 
1987-1988  3458579 
1988-1989  4176048

115 	
Compendium  of  Project  Reports 	
1989-1990  5132650 
1990-1991  6157753 
1991-1992  6616782 
1992-1993  8567098 
1993-1994  10488907 
1994-1995  12719229 
1995-1996  16302305 
1996-1997  18489354 
1997-1998  19823745 
1998-1999  17372692 
1999-2000  18857370 
2000-2001  17835033 
2001-2002  18322914 
2002-2003  21437562 
2003-2004  24777726 
2004-2005  30962009 
2005-2006  36469705 
2006-2007  46018006 
2007-2008  55275622 
2008-2009  61131148 
2009-2010  69718259 
2010-2011  82513335 
2011-2012  90520894 
2012-2013  100727950 
2013-2014  106511164 
2014-2015  116470249 
2015-2016  127327968 
2016-2017  136805049 
2017-2018  146697043 
2018-2019  153801928 
2019-2020  148574512 	
 	
               Source: www.mospi.gov.in	 	
5. Tools and Libraries 
 
      The analysis was done using Python Programming.  The technique used for time series 
analysis  was  ARIMA.  The  libraries  used  were  numpy,  pandas,  matplotlib, 
plotly.graph_objects,  seaborn,  sklearn,  statsmodels.tsa.stattools,  statsmodels.graphics. 
tsaplots, statsmodels.tsa.arima.model, scipy.stats, statsmodels.api. 
 
6.  Result with inference 	
The  GVA  of  Organized  Manufacturing  Sector  of  India  and  Kerala  for  40  years 	
from 1980-81 to 2019-20 were analysed using ARIMA model of time series analysis.  The 
model fitted for data of India and Kerala were used for in-sample and out-sample forecast.  
The  MAPE  was  used  to  measure  the  accuracy  of  the  results.    The  following  table  shows 
the results of the analysis.

116                  	Compendium  of  Project  Reports	 	
 	
In-sample Forecast of GVA of India 	
Year  Actual GVA 	
(in Rupee lakhs) 	
Predicted GVA (in 	
Rupee Lakhs) 	
2016-17  136805409  137428200 
2017-18  146697043  147741200 
2018-19  153801928  157994500 
2019-20  148574512  168264500 	
 	
Out-sample Forecast of GVA of India 	
Year  Predicted GVA   	
 (in Rupee Lakhs) 	
2020-21  178529800 
2021-22  188796500 
2022-23  199062800 
2023-24  209329200 	
 	
In-sample forecast of GVA of Kerala 	
   	
Year  Actual GVA 	
(in Rupee lakhs) 	
Predicted GVA 	
(in Rupee 
Lakhs) 	
2016-17  2236188  2004238 
2017-18  2183551  2146227 
2018-19  2337876  2444895 
2019-20  2252678  2479974 	
  	
Out-sample Forecast of GVA of Kerala 	 
 	
Year  Predicted GVA 	
(in Rupee Lakhs) 	
2020-21  2933417 
2021-22  3160725 
2022-23  3267800 
2023-24  3718621 	
 
 
 Conclusion  
 
         This  study  was  conducted  to  analyse  the  trend  of  GVA  of  India  and  Kerala  over  a 
period of 40years.  ARIMA method used is applicable only if the data is stationary (mean, 
varaiance  and  autocorrelation  remains  constant  through  time).    The  study  discusses  the 
changes  in  the  GVA  for  the  period  of  1980-81  to  2019-20.    The  results  of  the  study 
provides useful information for identifying the trend of GVA.   
 
References: 
 	
1.         	Fundamentals of Applied Statistics, S.C.Gupta, V.K.Kapoor, Sultan Chand & Sons      
Publication. 	
2.   Barzola_monteses, Monica Mite-Leon, Mayken Espinoza-Andaluz, Juan Gomez-Romero and 	
Wald  Fajardo,      et al.(2019).Time Series Analysis for Predicting Hydroelectric Power 
Production: The Ecuador Case. 	
3.   Habib Ahmed Elsayir, et al (2018), An Econometric Time Series GDP Model Analysis: 	
Statistical Evidences and Investigations.

117 	
Compendium  of  Project  Reports 
 	
Prediction on Live Births	
 	
Submitted  By 	
Sri. Prasanth B.R., 	Computer Supervisor

118                  	Compendium  of  Project  Reports	 	
 
Introduction	 	
  	
The  live  birth  data  for  the  Indian  state  of  Kerala  from  1962  to  2020  represents  the 	
number  of  live  births  that  occurred  in  the  state  each  year  during  this  period.  This  dataset 
can be analyzed to identify trends and patterns in the number of live births over time and to 
make predictions about future live birth rates in the state. 
   The term live birth refers to the delivery of a baby that is born with a heartbeat and 
breathing, indicating that the baby has viable life outside the womb. 
 
Objectives 
 	
The  objective  of  this  analysis  is  to  predict  the  number  of  live  births  in  Kerala  for 	
future  years  using  four  different  methods:  Simple  linear  regression,  Simple  linear 
regression  with  log,  ARIMA,  and  LSTM.  Simple  linear  regression  is  a  commonly  used 
statistical  method  that  can  help  us  understand  the  relationship  between  two  variables,  in 
this case, the year and the number of live births. By fitting a linear regression model to the 
data, we can estimate how much the number of live births changes for each additional year. 
In the Simple linear regression with log method, we will  take the logarithm of the number 
of  live  births  before  fitting  the  regression  model.  This  technique  can  help  to  stabilize  the 
variance of the data and improve the accuracy of the model. 	
ARIMA  stands  for  Autoregressive  Integrated  Moving  Average,  which  is  a  time 	
series  model  that  can  help  us  to  predict  future  values  based  on  past  observations.  This 
method takes into account both the trend and the seasonality of the data, which can help to 
improve the accuracy of the predictions. 	
Finally,  LSTM  stands  for  Long  Short-Term  Memory,  which  is  a  type  of  neural 	
network  model  that  is  particularly  suited  to  time  series  data.  LSTM  models  can  capture 
long-term  dependencies  in  the  data  and  are  often  used  in  applications  such  as  natural 
language processing and speech recognition. 	
By  comparing  the  performance  of  these  four  methods,  we  can  identify  which 	
technique is  most effective for predicting the number of live births in Kerala in the future. 
The  results  of  this  analysis  could  be  useful  for  policymakers  and  healthcare  professionals 
in  Kerala  who  need  to  plan  for  future  resource  allocation  and  healthcare  infrastructure 
based on the expected number of live births in the state. 
 
Methodolody and method used 
 
The objective of this analysis is to predict the live birth rates for the Indian state of Kerala 
from  1962  to  2020  using  four  different  models:  Simple  Linear  Regression,  Simple  Linear 
Regression  with  Logarithmic  Transformation,  Autoregressive  Integrated  Moving  Average 
(ARIMA), and Long Short-Term Memory (LSTM) neural networks.  
 
Methodology: 
 
1. 	Data  Cleaning:	  The  first  step  is  to  clean  the  dataset,  which  involves  checking  for  any 	
missing  or  erroneous  values  and  correcting  them  if  necessary.  The  dataset  appears  to  be 
complete, so no further cleaning is required. 
2. 	Exploratory  Data  Analysis:	 The second step is to conduct exploratory data analysis to 	
identify any trends or patterns in the data. This can be done by creating visualizations such

119 	
Compendium of Project Reports 
 
as  line  plots,  scatter  plots,  and  histograms.  This  step  helps  in  understanding  the 
characteristics  of  the  data  and  identifying  any  outliers  or  anomalies  that  may  need  to  be 
removed. 
3.	 Feature  Selection	: The third step is to select the relevant features for the model. In this 	
case, the only feature is the year, as it is the only independent variable in the dataset.  
4.  Model  Selection  and  Training:	  The  fourth  step  is  to  select  and  train  the  models.  The 	
four  models  selected  for  this  analysis  are  Simple  Linear  Regression,  Simple  Linear 
Regression with Logarithmic Transformation, ARIMA, and LSTM neural networks.   
    a	. 	Simple  Linear  Regression:	  This  model  assumes  a  linear  relationship  between  the 	
dependent  variable  (live  birth  rates)  and  the  independent  variable  (year).  The  model  is 
trained using the historical data and used to make predictions for future years. 
    b. Simple Linear Regression with Logarithmic Transformation:	 This model assumes 	
that the relationship between the dependent variable and independent variable is not linear, 
but can be transformed using a logarithmic function to make it linear. The model is trained 
on the transformed data and used to make predictions for future years. 
    c.	 ARIMA	:  This  model  is  a  time  series  analysis  technique  that  takes  into  account  the 	
autocorrelation and stationarity of the data. It involves identifying the order of differencing, 
the  order  of  autoregression,  and  the  order  of  moving  average.  The  model  is  trained  using 
historical data and used to make predictions for future years. 
    d	. LSTM:	  This  model  is  a  type  of  recurrent  neural  network  that  can  handle  sequence 	
data.  It  is  well-suited  for  time  series  analysis  as  it  can  capture  long-term  dependencies  in 
the data. The model is trained using historical data and used to make predictions for future 
years. 
5.	 Model  Evaluation:	  The  final  step  is  to  evaluate  the  performance  of  the  models.  This 	
can be done using metrics such as mean squared error, root mean squared error, and mean 
absolute error. The model with the lowest error is selected as the best model. 
Overall, this methodology involves cleaning the data, conducting exploratory data analysis, 
selecting relevant features, training and evaluating the models, and selecting the best model 
based on the evaluation metrics. 
Dataset 
 	
Year  LiveBirth 	 	Year  LiveBirth 	 	Year  LiveBirth 	
1962  395620  1982  560859  2002  581925 
1963  398244  1983  546192  2003  558369 
1964  424244  1984  548998  2004  563153 
1965  415109  1985  587966  2005  559082 
1966  428207  1986  567210  2006  537915 
1967  402151  1987  570283  2007  545154 
1968  415132  1988  574312  2008  535738 
1969  378716  1989  564066  2009  542154 
1970  305435  1990  568981  2010  546964 
1971  398159  1991  557697  2011  560268 
1972  468076  1992  523455  2012  550411

120                  	Compendium  of  Project  Reports	 	
 	
1973  466445 	 	1993  530470 	 	2013  536352 	
1974  471424  1994  533209  2014  534458 
1975  530425  1995  557787  2015  516013 
1976  529522  1996  586253  2016  496292 
1977  494888  1997  607727  2017  503597 
1978  498357  1998  591508  2018  488174 
1979  527839  1999  596948  2019  480113 
1980  532422  2000  593724  2020  446891 
1981  535746  2001  579063 	  	
 
Tools and libraries used 
 	
To  predict  live birth  data  for Kerala  from  1962  to  2020  using  different  techniques, 	
you can use several Python libraries and tools. Here are some common ones: 
1. Pandas: Pandas is a Python library used for data manipulation and analysis. You can use 
it to load and preprocess the live birth data for Kerala. 
2. NumPy: NumPy is a library used for numerical computing in Python. It provides support 
for  large,  multi-dimensional  arrays  and  matrices,  as  well  as  a  variety  of  mathematical 
functions. You can use it to perform mathematical operations on the live birth data. 
3.  Matplotlib:  Matplotlib  is  a  data  visualization  library  that  allows  you  to  create  various 
types  of  charts  and  graphs.  You  can  use  it  to  visualize  the  trends  and  patterns  in  the  live 
birth data. 
4. StatsModels: StatsModels is a Python library used for statistical modeling  and analysis. 
You  can  use  it  to  perform  linear  regression  analysis  and  time  series  analysis  on  the  live 
birth data. 
5.  Scikit-learn:  Scikit-learn  is  a  machine  learning  library  that  provides  various  algorithms 
for  data  analysis  and  modeling.  You  can  use  it  to  perform  machine  learning  techniques 
such as linear regression and LSTM on the live birth data. 
Once  you  have  loaded  and  preprocessed  the  live  birth  data  using  Pandas,  you  can  use 
StatsModels to perform simple linear regression and ARIMA analysis on the data. You can 
use  Scikit-learn  to  perform  simple  linear  regression  with  log  and  LSTM  analysis  on  the 
data.  You  can  use  Matplotlib  to  visualize  the  predictions  and  evaluate  the  performance  of 
each model. 
 
Result with inference 
 	
Diagram 01 represents the scattered diagram of live birth.

121 	
Compendium of Project Reports 
 	
The scatter plot would show the year on the x-axis and the number of live births on 	
the y-axis. By plotting the data points for each year, we can observe the trend of live births 
over the years. 	
Upon observing the scatter plot, we can see that the number of live births has been 	
fluctuating over the years. In the 1960s and 1970s, the number of live births was relatively 
low and gradually increased in the 1980s and 1990s. However, after 2000, the trend shows 
a gradual decline in the number of live births in Kerala. 	
 	
Diagram 02 Represents a Line Graph showing the relationship between LiveBirth 	
and Year	 	
 	 
The Line Graph also shows some years where there were significant spikes or drops in the 
number  of  live  births,  such  as  in  the  years  1972,  1975,  1996,  and  2002.  These  spikes  and 
drops  may  be  attributed  to  various  factors  such  as  changes  in  policies,  social  norms,  or 
economic conditions. 
Overall,  the  scatter  plot  and  Line  Graph  shows  live  births  in  Kerala  from  1962  to  2020 
shows a fluctuating trend with a recent decline in the number of live births. 
 
Prediction with Live birth dataset	  	
 	
1.	 Prediction based on Simple Linear Regression 
 
A  simple  linear  regression  model  was  used  with  a  training  data  set  of  0.7  and  a 
testing data set of 0.3. The Mean Squared Error (MSE) is a metric that measures the 
average  squared  difference  between  the  predicted  values  and  the  actual  values  in 
the test data set. A lower MSE value indicates better performance of the model.  In 
this  case,  the  MSE  value  is  2380213597,  which  is  a  relatively  high  value.  This 
suggests  that  the  model  may  not  be  accurately  predicting  the  dependent  variable 
based  on  the  independent  variable,  and  there  may  be  other  factors  that  are 
influencing the dependent variable that are not captured by the model. 
 
The  R  squared  value  is  a  metric  that  indicates  the  proportion  of  variance  in  the 
dependent variable that can be explained by the independent variable in the model. 
An  R  squared  value  of  0.4  indicates  that  40%  of  the  variance  in  the  dependent 
variable  can  be  explained  by  the  independent  variable  in  the  model.  This  value  is 
not very high, which suggests that there may be other variables that are influencing 
the dependent variable that are not included in the model.

122                  	Compendium  of  Project  Reports	 	
 	
Overall,  based  on  the  high  MSE  value  and  the  relatively  low  R  squared  value,  it 
appears that the simple linear regression model may not be the best fit for the data. 
It  may  be  worth  exploring  other  types  of  models  or  incorporating  additional 
variables to improve the accuracy and performance of the model. 
Based on this model  
 	
Diagram 04 	
 	
  2020  2021  2022 
Live Birth Prediction 
using SLR 	
558577	 	560232	 	561887	 	
 	
2.	 Prediction based on Simple Linear Regression with Log 
A  simple  linear  regression  model  with  log  transformation  was  used  for  prediction, 
with a training data set of 0.7 and a testing data set of 0.3. The performance of the 
model was evaluated using two metrics: Mean Squared Error (MSE) and R squared 
value. 
The  use  of  a  log  transformation  on  the  independent  variable  is  often  done  to  help 
normalize  the  data  and  reduce  the  influence  of  outliers.  The  performance  of  the 
model, as indicated by the MSE and R squared value, was as follows: 
The  Mean  Squared  Error  (MSE)  was  0.009,  which  is  a  very  low  value.  This 
indicates  that  the  model's  predictions  are  very  close  to  the  actual  values  in  the  test 
data set, suggesting that the model is performing well. 
The  R  squared  value  was  0.43,  which  indicates  that  43%  of  the  variance  in  the 
dependent variable can be explained by the independent variable in the model. This 
is higher than the R squared value of 0.4 obtained in the previous model, suggesting 
that  the  log  transformation  of  the  independent  variable  has  improved  the 
performance of the model. 
Overall,  the  simple  linear  regression  model  with  log  transformation  appears  to  be 
performing  well  based  on  the  low  MSE  and  relatively  high  R  squared  value. 
However,  it  is  important  to  keep  in  mind  that  there  may  still  be  other  variables  or 
factors  that  are  influencing  the  dependent  variable  that  are  not  captured  by  the 
model.  Further  analysis  and  exploration  may  be  needed  to  improve  the  accuracy 
and robustness of the model.

123 	
Compendium of Project Reports 
 	
 	
3.	 Prediction based on Simple Linear Regression with moving 
average 
 
A  simple  linear  regression  model  with  a  moving  average  was  used  for  prediction, 
with a training data set of 0.7 and a testing data set of 0.3. The performance of the 
model was evaluated using two metrics: Mean Squared Error (MSE) and R squared 
value. 
 
The  use  of  a  moving  average  on  the  independent  variable  is  often  done  to  smooth 
out  fluctuations  in  the  data  and  make  it  easier  to  identify  underlying  trends.  The 
performance  of  the  model,  as  indicated  by  the  MSE  and  R  squared  value,  was  as 
follows: 
 
The Mean Squared Error (MSE) was 1596213570, which is a relatively high value. 
This  indicates  that  the  model's  predictions  have  a  higher  degree  of  error  when 
compared to the actual values in the test data set. 
 
The  R  squared  value  was  0.19,  which  is  a  low  value.  This  indicates  that  the 
variation  in  the  dependent  variable  that  can  be  explained  by  the  independent 
variable in the model is only 19%. 
 
Overall,  the  performance  of  the  simple  linear  regression  model  with  moving 
average  appears  to  be  weaker  compared  to  the other  models  described  earlier.  The 
relatively  high  MSE  and  low  R  squared  value  suggest  that  the  model  is  not 
accurately  capturing  the  relationship  between  the  dependent  and  independent 
variables.  It  is  important  to  note  that  while  moving  average  can  be  helpful  in 
identifying  trends,  it  may  not  be  the  best  choice  of  transformation  for  all  datasets. 
Further  exploration  and  experimentation  with  different  transformation  techniques 
and  models  may  be  necessary  to  improve  the  accuracy  and  reliability  of  the 
predictions.

124                  	Compendium  of  Project  Reports	 	
 	
 	
4.	 Prediction based on ARIMA Auto Regression Model 
 	
The  ARIMA  (AutoRegressive  Integrated  Moving  Average)  model  is  a 	
commonly  used  time  series  model  that  is  suitable  for  forecasting  data  with  trend 
and/or  seasonal  patterns.  In  this  case,  the  ARIMA  model  has  been  used  for 
prediction with a training data set of 0.7 and a testing data set of 0.3. 
 	
The ARIMA model works by identifying patterns in the time series data and 	
using these patterns to make predictions about future values. This is typically done 
by  analyzing the autocorrelation and partial autocorrelation plots of the time series 
data to determine the appropriate parameters for the model. 
It is also worth noting that visual inspection of the graph generated by the ARIMA 
model  can  be  helpful  in  understanding  the  trend  and  seasonality  in  the  data,  and 
assessing the accuracy of the predictions. 	
Overall,  the  ARIMA  model  is  a  useful  tool  for  modeling  and  forecasting 	
time  series  data,  and  can  provide  valuable  insights  into  trends  and  patterns  in  the 
data.  However,  it  is  important  to  carefully  evaluate  the  performance  of  the  model 
using appropriate evaluation metrics and to consider alternative models if necessary.

125 	
Compendium of Project Reports 
 	
 	
  	
5.	 Prediction based on LSTM 
 	
LSTM  (Long  Short-Term  Memory)  is  a  type  of  neural  network  model  that 	
is commonly used for modeling and forecasting sequential data, such as time series 
data. In this case, the LSTM model has been trained using a training data set of 0.7 
and a testing data set of 0.3.  	
LSTM  models  are  particularly  well-suited  to  capturing  long-term 	
dependencies  in  time  series  data,  and  can  be  used  to  model  both  linear  and  non-
linear  relationships.  They  work  by  using  a  network  of  interconnected  nodes  that 
allow  information  to  flow  through  the  network  over  time,  and  are  capable  of 
remembering information from previous time steps. 
 	
However,  it  is  important  to  note  that  LSTM  models  can  be  complex  and 	
computationally  intensive  to  train,  and  may  require  significant  amounts  of  data  to 
achieve  good  performance.  Additionally,  the  selection  of  appropriate  hyper 
parameters,  such  as  the  number  of  nodes  and  the  learning  rate,  can  be  crucial  for 
obtaining good results. 
 	
Overall,  LSTM  models  are  a  powerful  tool  for  modeling  and  forecasting 	
time  series  data,  and  can  provide  valuable  insights  into  trends  and  patterns  in  the 
data.  However,  they  require  careful  selection  and  tuning  of  hyperparameters,  and 
their  performance  should  be  carefully  evaluated  using  appropriate  evaluation 
metrics.

126                  	Compendium  of  Project  Reports	 	
 	
 	 
Conclusion 
 	
Types of 
Model used 	
Training 
Data 	
Testing 
Data 	
Mean Squared 
Error 	
R square 
value 	Remarks 	
Simple Linear 
Regression 	
0.7  0.3 	2380213597  0.4  Got Prediction 	
based on input 	
Simple Linear 
Regression 
with Log 	
0.7  0.3 	0.009  0.43  Got Prediction 	
based on input 	
Simple Linear 
Regression 
with moving 	
average 	
0.7  0.3 	1596213570  0.19  Got Prediction 	
based on input 	
ARIMA Auto 
Regression 	
Model 	
0.7  0.3     	Graph 	
LSTM  0.7  0.3     	Graph 	
 
Reference 
 	
ï· 	Introduction  to  Time  Series  Forecasting  with  Python:  https://machinelearningmastery.com/time-
series-forecasting/ 	
ï· 	Simple Linear Regression: https://www.statisticssolutions.com/simple-linear-regression/ 	
ï· 	ARIMA: https://www.statisticshowto.com/arima/ 	
ï· 	LSTM:  https://machinelearningmastery.com/how-to-develop-lstm-models-for-time-series-
forecasting/

127 	
Compendium of Project Reports 
 	
 	
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 	
 	Submitted  By 	
Sri. Rajesh R, 	Statistical Assistant Grade I 	
 
 
 
 
 	
Analysing the correlation between 	
Birth weight and Maternal factors 	 	
A review of CRS data using Machine 	
Learning Approach

128                  	Compendium  of  Project  Reports	 	
 
1.Introduction 
  	 Civil  registration  is  the  process  of  recording  vital  events,  such  as  births,  deaths, 	
and  marriages,  and  maintaining  official  records  of  these  events  for  legal  and  statistical 
purposes.  In  the  Indian  state  of  Kerala,  the  civil  registration  system  is  an  essential 
component  of  the  state's  administrative  machinery,  responsible  for  recording  and 
maintaining  accurate  and  up-to-date  records  of  vital  events  occurring  in  the  state.	 	The 	
Civil  Registration  System  (CRS)  is  the  continuous,  regular  and  mandatory  registration  of 
important  events  (births  and  deaths)  of  a  population.	 	The  Births  and  Deaths  Registration 	
Act  1969  (RBD  Act)  mandated  the  registration  of  births,  deaths  and  stillbirths  in  India  on 
the  basis  of  place  of  occurrence.	 	The  Act  went  into  effect  in  the  state  along  with  several 	
other states on April 1, 1970 through an official  gazette notice dated March 21, 1970. The 
Act  requires  the  registration  of  births  and  deaths  and  the  deadline  for  notification  of  birth 
or death to the registrar is 21 days for registration. Registration can be extended beyond 21 
days under the terms of article 13 of the Law.	 	The reporting form for data is divided into 	
two  sections:  legal  and  statistical.  The  statistical  report  for  births,  deaths,  and  stillbirths  is 
created  using  data  that  is  electronically  registered  through  the  Sevana/ILGMS  software, 
developed  by  IKM  in  Kerala.	 	The  statistical  report  of  registered  births  and  deaths  is 	
prepared  by  Department  of  Economics  and  Statistics  (DES).  The  "Annual  Vital  Statistics 
Report"  published  annually  by  the  Department  is  the  main  publication  that  describes  the 
situation of CRS  in the State. This includes compiling data on births, deaths, infant deaths 
and stillbirths recorded from civil registries.	  	
 
 	Child birth is a complex process that can be influenced by  a wide range of factors, 	
including  the  child's  sex,  the  mother's  age  at  the  time  of  delivery,  the  number  of  children 
the mother has already, the duration of pregnancy, and the manner of delivery. One of the 
most  critical  indicators  of  a  child's  health  at  birth  is  their  birth  weight,  which  can  have 
long-term  implications  for  their  growth  and  development.  With  the  advent  of  machine 
learning techniques, it has become possible to analyze large datasets and identify complex 
patterns  and  relationships  between  different  variables.  In  this  context,  a  machine  learning 
approach can be used to identify any correlation between a child's birth weight and various 
maternal and delivery-related factors, such as the child's sex, the mother's age, the number 
of previous births, the duration of pregnancy, and the manner of delivery. Such an analysis 
can  provide  valuable  insights  into  the  factors  that  influence  a  child's  birth  weight  and 
enable  healthcare  providers  to  develop  targeted  interventions  to  improve  outcomes  for 
mothers  and  newborns.  By  leveraging  the  power  of  machine  learning,  this  approach  can 
help to identify previously  unknown relationships and patterns that may have a significant 
impact  on  maternal  and  child  health.  Ultimately,  this  can  lead  to  more  effective  and 
efficient healthcare interventions and improve outcomes for mothers and their newborns. 
 
2. OBJECTIVE 	
  To  assess  and  illustrate  the  relationship  between  a  child's  birth  weight  and  other 	
maternal  characteristics  including  childâs  sex,  the  mother's  age  at  delivery,  the  number  of 
children  the  mother  has  already,  gestational  age,  and  the  manner  of  delivery,  an  efficient 
machine learning technique should be developed. The analysis has been done using Python 
Programming.

129 	
Compendium of Project Reports 
 
METHODOLOGY AND METHOD USED  	
Trying  to  apply  multiple  regression  method  to  analyses  the  data.  Before  that  we 	
have to make the data suitable for the study Which typically involves following steps. 	
 	
1. 	Data  collection:	  Collect  the  dataset  that  includes  the  predictor  variables  (also  called 	
independent variables or features) and the target variable (also called dependent variable or 
response variable)  
The dataset used for the study is Civil Registration System data - Statistical part from 2008 
to 2019.	 Data consist of a total of 629272 (more than 62 lakhs) records. 	  
2.  Data  cleaning  and preprocessing: 	Perform data cleaning and preprocessing tasks such 	
as removing missing values, handling outliers, and scaling the data. 	
Here  in  our  data  by  checking  the  data  with  outliers  and  when  visualizing  it  found 	
that  there are certain  outliers  which  affect  the quality  of the data  and  our  study  so  have to 
remove it first. 	
 	
 
By  analysing  the  graph  it  is  clear  that  weight  of  the child  versus  pregnancy  period 	
graph has outliers on weight  as 9kg (which is actually the code used for unknown weight) 
and  pregnancy    period  as  99  months  (which  is  actually  the  the  code  used  for  unknown 
months) any way have to remove such outliers.

130                  	Compendium  of  Project  Reports	 	
 	
From  this  data,  It  is  clear  that  weight  of  the  child  versus  number  of  children  the 	
mother  have  graph    has  outliers  on  number  of  children  as  99  (which  is  actully  the  code 
used  for  unspecified  number  of  children  the  mother  had)  any  way  have  to  remove  such 
outliers	. 	
 	
By  analysing  the 	above  data,  It  is  clear  that  weight  of  the  child  versus  Delivery  type 	
graph  has  some  outliers  on  delivery  type  as  9(which  is  actualy  the  code  used  for 
unspecified delivery type )any way have to remove such outliers. 
       In  order  to  remove  outliers,  run  inter  quartile  range  in  python  and  that  removes  the 
outliers stated above and get a clean data and since we used published dataset for the study 
so there is no scope for missing values on all columns. 
3. Splitting the data	: Split the dataset into training and testing sets using a holdout method 	
or cross-validation. 
         	In order to train the system we split data set into training dataset and testing dataset. 	
Commonly  we  split  the  data  into  80:20  ratios.  But  here,  since  the  dataset  was  very  large 
(contains more than 62 lakhs of data) we split the data set into 99:1 ratio (only for plotting). 
4.	 	Building  the  model	:  Build  a  multilinear  regression  model  using  a  suitable  Python 	
library such as scikit-learn or statsmodels etc.  
5. 	Mode	l training	: Train the model on the training set. 	
       	Train  the  system  with  training  dataset  and  for  the  evaluation  we  apply  that  on  testing 	
dataset and check the accuracy.  
6.  Model  evaluation	:  Evaluate  the  model's  performance  on  the  testing  set  using 	
appropriate evaluation metrics such as mean squared error (MSE), root mean squared 	error 	
(RMSE), and R-squared. 
4.  DATASET USED  
          	 	
         The  dataset  used  for  the  study  was  Civil  Registration  System  â  Birth  data  from  the 
year  2008  to  the  year  2019.  The  data  consist  of  a  total  of  629272  (more  than  62  lakhs) 
records. Dataset has column headings as

131 	
Compendium of Project Reports 
 	
REG_YEAR                    1093403 non-null  int64 
1   SEX                                   1093402 non-null  float64 
2   MOTHERS_AGE_DELIVERY  1093400 non-null  float64 
3   NO_CHILDREN           1093396 non-null  float64 
4   PREAGNANCY_PERIOD     1093257 non-null  float64 
5   DELIVERY_TYPE         1093393 non-null  float64 
6   WEIGHT_CHILD          1093393 non-null  float64  	
The row data is of the form 	
 	
REG_ 
YEAR 	
SEX  MOTHERS_ 	
AGE_ 	
DELIVERY 	
NO_ 	
CHILDREN	 	
PREAGNANCY_	 	
PERIOD 	
DELIVERY_ 	
TYPE 	
WEIGHT_ 
CHILD 	
2019  2  28  1  39  2  3.67 
2019  2  21  1  40  1  2.77 
2019  1  28  4  37  2  2.59 
2019  2  33  3  38  2  3 
2019  1  31  2  38  2  3.27 
2019  2  30  3  39  2  2.51 
2019  2  31  2  38  2  3 
2019  2  27  4  40  2  3.7 
2019  1  28  3  38  2  4.05 
2019  1  19  1  40  1  2.7 
2019  2  28  2  39  2  3.37 
2019  1  19  1  38  1  2.53 
2019  1  26  1  39  1  3.28 
2019  1  25  2  40  1  4 
2019  1  27  2  38  1  2.95 
2019  2  29  2  39  2  3.5 
2019  1  29  2  40  2  2.75 
2019  2  25  1  40  2  3 
2019  1  36  3  40  2  3 
2019  2  31  2  37  2  2.8 
2019  2  27  1  39  2  3.5 
2019  2  26  2  39  2  2.4 
2019  2  32  2  40  2  3 
2019  2  23  1  40  2  3.2 
2019  2  26  1  38  1  3 
2019  2  26  1  40  2  3.25 
2019  2  24  1  39  1  2.9 
2019  1  25  2  38  2  3.05 
2019  2  30  2  36  2  3.2 
2019  1  25  1  36  1  3.04 
2019  1  27  1  39  1  3.25 
2019  2  27  1  40  2  2.75 
2019  1  26  2  38  2  3.5 
2019  2  31  4  38  1  3 
2019  1  33  2  39  2  2.47 
2019  1  28  2  37  1  2.6 
2019  2  27  1  39  1  2.87 
2019  1  33  3  37  1  3 
2019  2  29  2  40  1  3.11 
2019  2  31  1  37  2  2.84 
2019  2  21  1  38  1  3.22 
2019  2  31  4  39  2  3.65 
2019  2  20  1  40  2  2.5 
2019  2  28  2  39  1  3 
2019  2  35  2  39  2  3.3 
2019  1  23  1  40  2  2.5 
2019  1  29  2  37  2  3 
2019  1  25  1  38  2  2.75 
2019  2  23  1  38  1  2.4 
2019  1  23  2  38  2  3

132                  	Compendium  of  Project  Reports	 	
 	
And the data set after cleaning is of the form 	
 	
  Reg-Year  Sex  Months- Age -	
Delivery 	
Pregnancy 	
period 	
Delivery Type  Weight 	
Child 	
6292723  2019  2  23  1  37.0  2 
6292724  2019  1  36  3  40.0  2 
6292725  2019  2  19  1  39.0  2 
6292726  2019  2  29  1  39.0  2 
  	
In order to apply multi linear regression model. We consider weight of the child as 	
dependent  variable  and  Sex  of  the  child,  Motherâs  age  at  delivery,  number  of  children, 
pregnancy period, delivery type as independent variables.	 	
 
 	
 	
From heat map of correlation it is clear that sex of the child ,Registration year have 	
no correlation with the weight of the child so sex  of the child does not consider for further 
study.  Now  the  weight  of  the  child  as  dependent  variable  and  Motherâs  age  at  delivery, 
number of children, pregnancy period, delivery type as independent variables. And now fit 
the data into multi linear regression model	. 	
5.TOOLS AND LIBRARIES USED:  	
In  python,  there  are  several  libraries  available  for  machine  learning  using  multi-
linear regression, and here we use the following	. 	
 
3.1	 NumPy:	  A  fundamental  scientific  computing  library  for  Python  that  provides 	
support  for  large  multi-dimensional  arrays  and  matrices,  and  a  wide  range  of 
mathematical functions. 
 
3.2	 Pandas:	  A  data  analysis  library  that  provides  data  structures  and  tools  for 	
manipulating and analyzing numerical tables and time series data. 
 
3.3  Statsmodels: 	A  statistical  library  that  provides  tools  for  exploring  data, 	
estimating statistical models, and performing statistical tests.   
3.4 Seaborn

133 	
Compendium of Project Reports 
 	
Data 	visualization	:  Seaborn  provides  a  wide  range  of  functions  for      visualizing 	
data,  such  as  scatter  plots,  histograms,  box  plots,  heatmaps,  and  more.  These 
functions  can  be  used  to  explore  the  relationships  between  variables,  identify 
patterns in the data, and detect outliers or anomalies. 
 Feature  engineering:	 For feature engineering, which involves transforming 	
the  raw  data  into  a  format  that  is  suitable  for  machine  learning  algorithms? 
Seaborn  provides  functions  for  scaling,  encoding,  and  transforming  data, 
which can help to improve the performance of machine learning models. 
 Model  evaluation	:  For evaluating the performance of machine learning models. 	
For  example,  Seaborn  provides  functions  for  generating  confusion  matrices, 
classification  reports,  and  ROC  curves,  which  can  be  used  to  assess  the  accuracy, 
precision, recall, and other performance metrics of a model. 
3.5 Matplotlib.pyplot 	
 It is a useful tool for visualizing and analyzing the data and results 	
3.6 Sklearn.linear_model	: 	
Is  a  module  in  Scikit-learn  library  in  python.  The  linear  regression  class 
in  sklearn.linear_model  can  be  used  to  fit  a  linear  regression  model  to  a 
dataset and make predictions on new data.  	
6.  	RESULT AND INFERENCE: 	
  
 Calculated  the mean squared error value	  	
 
 Mean Absolute Error:  0.3503968182783409 
 	
The  value  .035  represents  the  average  absolute  difference  between  the 	
predicted and actual values of the target variable, the closer MAE value to zero the 
better the model making predictions. 
Based on our model we plot the graph

134                  	Compendium  of  Project  Reports	 	
 
 
 
             The prediction made is : 
             This the actual data  in the data set  	
 	
             And our predicted value is  
             predict_final=model.predict([[22.0,1.0,39.0,1.0]]) 
 
               	OUTPUT IS 3.86kg 	
 
CONCLUSION: 
 	
There  is  a  difference  between  actual  and  predicted  value  (3.43-3.86=  -  .43  kg). 	
Hence  more  studies  are  needed  to  identify  an  accurate  model  to  fit  the  data  and  make 
predictions and may be other factors like nutritionâs etc. are to be considered to predict the 
weight of the child in a better manner. 
REFERENCE: 	
ï· 	Journal  of  Obstetrics  and  Gynaecology,  used  decision  tree  and  random  forest 
algorithms  to  analyse  data  from  over  33,000  deliveries: 
https://journals.sbmu.ac.ir/aab/article/download/15412/13220 	
ï· 	Early  Prediction  of  Weight  at  Birth  Using  Support  Vector  Regression:       
https://link.springer.com/chapter/10.1007/978-3-030-30648-9_5 	
ï· 	Sex  differences  in  outcomes  of  very  low  birth  weight  infants: 
https://fn.bmj.com/content/83/3/F182

135 	
Compendium of Project Reports 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
 	
 	Submitted By 	
Sri. Preeth V.S	., Deputy Director (Nosologist) 	
 
 
 
 
  
 
 
 
 
 
 
 
 
 	
A Machine Learning Approach to 	
classification of cause of death

136                  	Compendium  of  Project  Reports	 	
 
 
 
 	
1.	 Introduction 
 	
Reliable  cause-specific  mortality  statistics  is  required  on  a  regular  basis  by 	
Administrators, Policy Planners, Researchers and other Professionals for evidence-
based decision-making with regard to resource allocation, monitoring of indicators, 
identifying  the  priorities  for  programs  and  other  related  activities  in  the  area  of 
Public Health.  Keeping  this  in  view,  the  scheme of Medical  Certification  of  Cause 
of  Death  (MCCD)  was  introduced  in  the  country  under  the  provisions  of  the 
Registration  of  Births  and  Deaths  (RBD)  Act,  1969.  In  Kerala  the  scheme  is 
presently  implemented  only  in  four  Corporations  viz.  â  Thiruvananthapuram, 
Kollam, Ernakulum and Kozhikode and in Alappuzha Municipality. Department of 
Economics  and  Statistics  (DES)  cross  tabulates  the  data  on  medically  certified 
cause of deaths reported from the above-mentioned 5 centres in conformity with the 
International Classification of Diseases (ICD) by age and sex of the deceased. 	
The necessary data is collected in the prescribed forms (Form 4 for Hospital 	
deaths  and  Form  4A  for  Non-institutional  deaths).  Both  these  forms  have  been 
designed by the World Health Organization (WHO). The forms are filled-up by the 
medical  professionals  attending  to  the  deceased  at  the  time  of  terminal  illness. 
Thereafter,  these  forms  are  to  be  sent  to  the  concerned  Registrars  of  Births  and 
Deaths  for  onward  transmission  to  the  Chief  Registrar  Office  for  tabulation  as  per 
the  National  List  of  Causes  of  Death  based  on  Tenth  Revision  of  International 
Classification  of Disease  (ICD-10).  The  coding  is  done by  Deputy  Health  Officers 
(DHO)  in  the  corporation/Municipality  offices.  Nosologist  in  DES  cross  tabulates 
the  data  by  cause  of  death,  age and  sex  in  excel  format  and  prepares  consolidation 
statements  in  prescribed  format.  The  States/UTs  subsequently  send  it  to  the Office 
of RGI in the form of Statistical Table-11 for consolidation at the National level. 	 	
2.	  Objective 	
To create  a model that  will be able to recognize and determine handwritten 	
characters  from  scanned  images  of  Medical  certification  of  Cause  of  death  forms 
filled  by  doctors  and  to  classify  the  cause  of  deaths  as  per  international 
classification of Diseases (ICD-10). 	
 	
3.	 Methodology 	
3.1.	    	Data  collection	:  This  step  involves  collecting  a  dataset  of  scanned  handwritten 	
death  certificates.  The  dataset  should  be  labelled  with  the  cause  of  death  for  each 
certificate.  This  means  that  each  certificate  should  be  associated  with  a  specific 
category of cause of death, such as heart disease, cancer, or respiratory illness. 	
3.2.	  Text  extraction	:  After  collecting  the  dataset,  the  next  step  is  to  use  Optical 	
Character  Recognition  (OCR)  or  handwriting  recognition  methods  to  extract  the 
text  from  the  scanned  documents.  The  extracted  text  should  be  cleaned  and  pre-
processed to remove any inaccuracies or inconsistencies. 	
3.3.	 Feature  extraction	:  The  next  step  is  to  extract  relevant  features  from  the  text  that 	
can  be  used  for  classification.  These  features  may  include  words,  phrases,  or  other

137 	
Compendium of Project Reports 
 	
information  related  to  the  cause  of  death.  For  example,  if  a  death  certificate 
mentions  the  phrase  "heart  attack",  this  can  be  used  as  a  feature  to  classify  the 
cause of death as heart disease. 	
3.4.	 Model  training	:  After  feature  extraction,  the  next  step  is  to  train  a  machine 	
learning  model  on  the  labelled  dataset  using  the  extracted  features.  This  can  be 
done  using  supervised  learning  techniques  such  as  Support  Vector  Machines, 
Random  Forest,  Naive  Bayes,  or  Decision  Trees.  The  goal  is  to  develop  a  model 
that can accurately classify the cause of death based on the extracted features. 	
3.5.	 Model  evaluation	:  Once  the  model  is  trained,  the  next  step  is  to  evaluate  its 	
performance  using  appropriate  metrics  such  as  accuracy,  precision,  recall,  and  F1-
score. This helps to assess the effectiveness of the model in correctly classifying the 
cause of death from the scanned handwritten death certificates. 	
3.6.	 Model  optimization	:  If  the  model  is  not  performing  well,  it  may  be  necessary  to 	
perform  model  optimization  by  adjusting  the  parameters  or  using  different  feature 
extraction  methods.  This  helps  to  improve  the  accuracy  of  the  model  and  make  it 
more effective in classifying the cause of death. 	
3.7.	 Model  deployment	:  Once  the  model  is  trained  and  optimized,  it  can  be  deployed 	
for  use  in  classifying  the  cause  of  death  from  new,  unseen  scanned  handwritten 
death certificates. This can be done by integrating the model into an  application or 
system  that  can  automatically  classify  the  cause  of  death  based  on  the  scanned 
handwritten death certificate input. 	
4.	 Models used 	
4.1.	 Support  Vector Classifier ( SVC) 	
The  Support  Vector  Classifier  (SVC)  is  a  popular  supervised  learning 	
algorithm  used  in  machine  learning  for  classification  tasks.  It  is  a  type  of  Support 
Vector  Machine  (SVM)  algorithm  that  is  effective  in  solving  both  linear  and  non-
linear classification problems.. 
 	
The  SVC  algorithm  works  by 	
finding  the  hyperplane  that  maximally 
separates  the  different  classes  in  the 
input data. This hyperplane is determined 
by  the  support  vectors,  which  are  the 
data points that lie closest to the decision 
boundary.  The  SVC  algorithm  tries  to 
find  the  hyperplane  that  maximizes  the 
margin,  which  is  the  distance  between 
the support vectors and the decision boundary. By maximizing the margin, the SVC 
algorithm  is  able  to  minimize  the  generalization  error,  which  is  the  error  rate  on 
new, unseen data. 	
One  of  the  benefits  of  using  the  SVC  algorithm  is  that  it  is  effective  in 	
handling  high-dimensional  data,  which  makes  it  suitable  for  use  in  text 
classification,  image  recognition,  and  other  types  of  classification  tasks. 
Additionally,  SVC  is  also  able  to  handle  both  linearly  separable  and  non-linearly 
separable  datasets  by  using  different  types  of  kernels.  Some  common  kernels  used

138                  	Compendium  of  Project  Reports	 	
 	
with  the  SVC  algorithm  include  linear,  polynomial,  radial  basis  function  (RBF), 
and sigmoid kernels. 	
In terms of implementation, the SVC algorithm can be used in Python using 	
the  scikit-learn  library.  The  SVC  class  in  scikit-learn  provides  various 
hyperparameters  that  can  be  tuned  to  improve  the  performance  of  the  algorithm, 
such as the kernel type, regularization parameter, and gamma parameter. 	
4.2.	 NaÃ¯ve Bayes Classifier 
 	
The  NaÃ¯ve  Bayes  Classifier  is  a 	
simple  yet  powerful  algorithm  used  in 
machine  learning  for  classification  tasks. 
It  is  based  on  the  Bayes  theorem  of 
probability theory  and is commonly used 
in  text  classification,  sentiment  analysis, 
spam  filtering,  and  other  applications 
where  data  is  represented  as  a  collection 
of  features.The  NaÃ¯ve  Bayes  Classifier 
works by assuming that the presence of a 
particular  feature  in  a  class  is 
independent  of  the  presence  of  other 
features. This is known as the "naive" assumption, which simplifies the probability 
calculations  and  makes  the  algorithm  computationally  efficient.  Based  on  this 
assumption,  the  NaÃ¯ve  Bayes  Classifier  calculates  the  probability  of  a  data  point 
belonging  to  a  particular  class  based  on  the  conditional  probability  of  each  feature 
given  the  class.  The  class  with  the  highest  probability  is  then  assigned  to  the  data 
point.	 	
One of the benefits of using the NaÃ¯ve Bayes Classifier is its simplicity and 	
speed.  The  algorithm  requires  only  a  small  amount  of  training  data  and  can  be 
trained  quickly  even  on  large  datasets.  Additionally,  the  NaÃ¯ve  Bayes  Classifier  is 
robust to irrelevant features, meaning that it can effectively classify data even when 
there are features that are not useful for classification. 	
There  are  three  types  of  NaÃ¯ve  Bayes  Classifier:  Bernoulli,  Multinomial, 	
and Gaussian. The Bernoulli NaÃ¯ve Bayes Classifier is used for binary classification 
problems  where  the  features  are  binary  variables.  The  Multinomial  NaÃ¯ve  Bayes 
Classifier  is  used  for  multi-class  classification  problems  where  the  features  are 
counts of occurrences. The Gaussian NaÃ¯ve Bayes Classifier is used for continuous 
features that are assumed to follow a Gaussian (normal) distribution. 	
In  terms  of  implementation,  the  NaÃ¯ve  Bayes  Classifier  can  be  used  in 	
Python  using  the  scikit-learn  library.  The  Gaussian  NaÃ¯ve  Bayes  Classifier  is 
available  in  scikit-learn's  GaussianNB  class,  while  the  Multinomial  NaÃ¯ve  Bayes 
Classifier and Bernoulli NaÃ¯ve Bayes Classifier are available in the MultinomialNB 
and BernoulliNB classes, respectively.

139 	
Compendium of Project Reports 
 	
 	
4.3.	  Decision Tree Classifier 
 	
A  decision  tree  classifier  is  a  type  of  machine  learning  algorithm  that  is 	
commonly used for classification tasks. It builds a tree-like model of decisions and 
their possible consequences, based on a set of training data. 	
 	
 	
 	
 
The  decision  tree  starts  with  a  root  node  and  branches  out  into  different 	
paths,  with  each  path  corresponding  to  a  possible  decision  based  on  a  specific 
feature  or  attribute.  Each  decision  or  split  in  the  tree  is  made  based  on  the  feature 
that provides the most information gain, meaning it separates the data into the most 
distinct groups. 	
The  process  continues  recursively  until  a  stopping  criterion  is  met,  such  as 	
when  all  instances  in  a  branch  belong  to  the  same class  or  when  the  tree  reaches  a 
maximum  depth.  Once  the tree  is  built,  it  can  be used  to  classify  new  instances  by 
traversing  down  the  tree  from  the  root  to  a  leaf  node,  which  corresponds  to  a 
particular class label. 	
Decision  trees  have  several  advantages,  including  their  interpretability  and 	
ability  to  handle  both  numerical  and  categorical  data.  However,  they  may  suffer 
from  over  fitting  and  instability,  and  may  not  perform  well  on  datasets  with  high 
dimensionality  or  noisy  data.  Ensemble  methods,  such  as  random  forests  and 
gradient boosting, can help overcome some of these limitations. 	
 	
4.4.	 Random Forest Classifier  	
Random Forest Classifier is an ensemble machine learning algorithm that is 	
used for classification tasks.  It is  an extension of the decision tree algorithm  and is 
based on the concept of building multiple decision trees and combining their results 
to make predictions.

140                  	Compendium  of  Project  Reports	 	
 	
 
 
 
 
 
 
 
 
 
 	
 
 
 
 
 
 
 
 
 
 
The  algorithm  works  by  randomly  selecting  a  subset  of  features  and 	
building  decision  trees  based  on  these  subsets.  This  process  is  repeated  multiple 
times  to  create  a  forest  of decision  trees.  During  training,  each  decision  tree  in  the 
forest  is  trained  on  a  random  subset  of  the  training  data,  with  replacement  (i.e., 
bootstrap  samples).  The  final  prediction  is  made  by  aggregating  the  predictions  of 
all  the  decision  trees  in  the  forest,  either  by  taking  the  majority  vote  or  averaging 
the predictions.  	
The  use  of  multiple  decision  trees  in  a  random  forest  has  several 	
advantages.  First,  it  reduces  the  risk  of  over  fitting,  as  the  average  prediction  of 
many  trees  is  often  more  stable  and  accurate  than  the  prediction  of  a  single  tree. 
Second,  the  algorithm  can  handle  both  numerical  and  categorical  data  and  can 
identify important features for classification. Third, it can deal with missing values 
in the dataset. 	
Random  forest  classifiers  have  been  shown  to  be  effective  in  a  wide  range 	
of  classification  problems,  including  text  classification,  image  classification,  and 
bioinformatics.  However,  they  can  be  computationally  expensive  and  may  require 
tuning of hyper parameters such as the number of trees in the forest and the size of 
the subsets used for feature selection. 	
 
 	
4.5. 	Multilayer Perceptron (MLP) classifier	  	
 	
Multilayer  Perceptron  (MLP)  classifier  is  a  type  of artificial  neural  network  that  is 	
used  for  classification  tasks.  It  is  a  feedforward  neural  network  that  consists  of 
multiple  layers  of  nodes  or  neurons,  with  each  layer  connected  to  the  next  in  a 
sequence.  The  input  layer  of  the  MLP  receives  the  input  data,  which  is  then  passed 
through  one  or  more  hidden  layers,  each  of  which  consists  of  multiple  neurons.  The 
neurons  in  each  hidden  layer are  fully  connected  to  the  neurons  in  the  previous  layer,

141 	
Compendium of Project Reports 
 	
and  each  neuron  in  a  hidden  layer  computes  a  weighted  sum  of  the  inputs  from  the 
previous  layer  and  applies  an 
activation function to produce its 
output.  The  output  layer 
produces  the  final  classification 
result.	  	
 	
The  MLP  classifier  is  trained 	
using  a  supervised  learning 
algorithm,  such  as  backpropagation. 
During  training,  the  weights  and 
biases  of  the  neurons  are  adjusted 
iteratively  to  minimize  the  error 
between  the  predicted  output  and  the  actual  output.  The  error  is  measured  using  a  loss 
function, such as cross-entropy or mean squared error. 	
   MLP  classifiers  are  widely  used  in  many  applications,  including  image 	
recognition,  speech  recognition,  and  natural  language  processing.  They  are  powerful  and 
flexible  models  that  can  learn  complex  patterns  in  the  input  data.  However,  they  can  be 
computationally expensive and may require tuning of hyperparameters such as the number 
of hidden layers, the number of neurons in each layer, and the learning rate. 	
 	
5. Data sets used 	
A total of 100 scanned images of MCCD forms for the year 2022 were gathered to 	
extract  text  information.  Additionally,  the  MCCD  data  for  the  year  2022  in  Excel  format 
consists  of  17000  cases  was  utilized  to  train  the  machine  learning  models  mentioned 
above. 	
 	
6. Tools and Libraries used 
 
6.1  Tesseract  OCR-  Tesseract  OCR  is  an  open-source  optical  character  recognition 	
(OCR) engine developed by Google. OCR is a technology that allows machines to 
recognize printed or handwritten text within an image and convert it into machine-
readable  text.  In  the  project,  the  Pytesseract  library  is  utilized  to  integrate  with 
Tesseract. 	
6.2   Pandas  - for data manipulation and analysis 
6.3. scikit-learn (sklearn for short)- is a free and open-source machine learning library for 	
the  Python  programming  language.  It  provides  tools  for  data  analysis  and 
preprocessing,  classification,  regression,  clustering,  model  selection  and 
evaluation, and more. The library is built on top of NumPy, SciPy, and Matplotlib, 
which are popular Python libraries for scientific computing and data visualization. 	
6.4.        TfidfVectorizer  -  for  converting  text  data  into  numerical  feature  vectors  using  the 	
term frequency-inverse document frequency (TF-IDF) approach 	
6.5.      SVC  (Support  Vector  Classification)  -  for  training  the  SVM  (Support  Vector 	
Machine) classifier model 	
6.6.    MultinomialNB -is a class in the sklearn.naive_bayes module of scikit-learn library 	
in  Python,  which  implements  the  Naive  Bayes  algorithm  for  multinomially 
distributed data.

142                  	Compendium  of  Project  Reports	 	
 
6.7.      DecisionTreeClassifier- a class for creating Decision tree classifier 
6.8.    MLPClassifier  -  for  creating  a  multi-layer  perceptron  (MLP)  neural  network 	
classifier 	
6.9.      Pipeline  -  for  chaining  together  multiple  machine  learning  steps  into  a  single 	
workflow 	
6.10.    train_test_split - for splitting the dataset into training and testing sets 
6.11.   GridSearchCV - for performing hyperparameter tuning of the SVM classifier using 	
grid search 	
6.12.   accuracy_score - for calculating the accuracy of the classifier 
6.13.   precision_score - for calculating the precision of the classifier 
6.14.   recall_score - for calculating the recall of the classifier 
6.15.   f1_score - for calculating the F1-score of the classifier 
6.16.    confusion_matrix  -  for creating  a  confusion  matrix  to  evaluate  the performance of 	
the classifier 	
6.17.  classification_report  -  for  generating  a  report  of  the  classification  performance, 	
including precision, recall, and F1-score 	
6.18.   seaborn - for creating visualizations, such as heatmaps 
6.19.   matplotlib.pyplot - for creating visualizations, such as line plots and scatter plots. 
6.20.   RandomForestClassifier: a class for creating a random forest classifier, which is an 	
ensemble of decision trees. 	
7.  Results 
7.1.  Text Extraction 	
Tesseract  OCR  was  not  successful  in  correctly  recognizing  the  text  owing  to  the 	
unfamiliarity of the engine with the handwritten characters of doctors. One of such result is 
shown  below.  Although  Google  Cloud  API  Vision  was  utilized  to  enhance  the  results,  it 
too failed to deliver accurate outcomes. 
 
 
 
 
 
 
 
 
 
 
 
 
 	
7.2. Evaluation of Models 	
 	
As  the  text  extraction  process  failed,  the  models  (SVC  classifier,  Multinomial  NB 	
classifier,  Decision  Tree  Classifier,  Random  Forest  Classifier,  and  MLP  classifier)  were 
trained  using  MCCD  data  in  an  Excel  format.  To  determine  the  optimal  model  for  the 
classification  problem,  the  accuracy,  F1-score,  recall,  and  precision  of  the  models  under 
consideration, were computed. Below are brief definitions of these terms.

143 	
Compendium of Project Reports 
 	
Accuracy:	  This  is  the  proportion  of  correctly  classified  instances  out  of  all 	
instances  in  the  dataset.  It  is  calculated  by  dividing  the  number  of  correctly  classified 
instances by the total number of instances. 	
F1-score:	  This  is  a  weighted  average  of  the  precision  and  recall  metrics.  It  is  a 	
measure of a model's accuracy that considers both the precision and recall. The F1-score is 
often  used  in  cases  where  there  is  an  imbalance  in  the  class  distribution  (i.e.,  when  one 
class is much more prevalent than the other). 	
Recall:	 This is the proportion of positive instances that were correctly classified by 	
the model out of all actual positive instances. It is also known as sensitivity or true positive 
rate. 	
Precision:	 This is the proportion of positive instances that were correctly classified 	
by  the  model  out  of  all  instances  classified  as  positive  by  the  model.  It  is  also  known  as 
positive predictive value. 	
 	
Table 7.2.1 Accuracy of the Models under consideration 
 
 	
Table  7.2.1  depicts  the  accuracy,  F1-score,  recall,  and  precision  of  the  models 	
under  consideration.	 it  seems  that  the  MLP  model  has  the  highest  F1-score  of  98.13%, 	
which combines both precision and recall into a single metric. The MLP model also has the 
highest  accuracy  of  98.43%,  indicating  that  it  correctly  classified  a high  percentage  of  the 
samples. 
  Although  the  SVC,  Decision  Tree,  and  Random  Forest  models  also  have  high 
accuracy,  precision,  and  recall  values,  their  F1-scores  are  slightly  lower  than  the  MLP 
model.  This  indicates  that  while  they  correctly  classified  a  high  percentage  of  samples, 
they  may  have  had  slightly  more  false  positives  or  false  negatives  than  the  MLP  model.
  The  MultinomialNB  model  has  a  significantly  lower  accuracy,  precision,  recall, 
and  F1-score  compared  to  the  other  models,  indicating  that  it  may  not  be  the  best  choice 
for this classification problem. 
  Overall, based on the provided table, the MLP model appears to be the best choice 
for  this  classification  problem,  as  it  has  the  highest  F1-score  and  accuracy  among  the 
models under consideration. 
 
7.3 Model optimization 
  Even though all the models have achieved high accuracy, we attempted to optimize 
theMLP  model  by  fine-tuning  its  hyperparameters.  Below  are  the  specific  parameters  that 
were fine-tuned for the MLP classifier. 
 	
Model   accuracy  precision  recall  F1-score 
SVC	 	98.36  98.11  98.36  98.06 	
Multinomial NB	 	96.36  94.98  96.35  95.4 	
Decision Tree	 	98.36  98.11  98.36  98.05 	
Random Forest	 	98.36  98.11  98.36  98.05 	
MLP	 	98.43  98.18  98.43  98.13

144                  	Compendium  of  Project  Reports	 	
 
Hidden layers	: The number of hidden layers in the neural network. Each hidden layer is a 	
layer  of  neurons  between  the  input  and  output  layers.  By  default,  MLPClassifier  uses  a 
single hidden layer with 100 neurons. 
 
Alpha	:  A  regularization  parameter  that  controls  the  strength  of  L2  regularization  on  the 	
weights of the neural network. L2 regularization helps prevent overfitting by shrinking the 
weights towards zero. By default, alpha is set to 0.0001. 
 
Solver:	  The  algorithm  used  to  optimize  the  weights  and  biases  of  the  neural  network 	
during training. The possible values are 'lbfgs', 'sgd', and 'adam'. By default, MLPClassifier 
uses 'adam'. 
 
Verbose	:  Controls  the amount  of  information  printed  during  training.  The  possible  values 	
are True or False. If set to True, the training progress is printed to the console. By default, 
verbose is set to False. 
 
Learning_rate_init:	  The  initial  learning  rate  used  by  the  optimizer.  This  determines  the 	
step size taken during weight updates. By default, learning_rate_init is set to 0.001. 
 
Table 7.3.1 Accuracy of the model for different values of hyper parameters. 	
 	
Trial  Hidden layers  alpha  verbose  learning_rate_init  solver  Accuracy 
1 	(10,10)  0.0001  10  0.1  sgd  98.36 	
2 	(50,50)  0.0001  10  0.1  sgd  98.36 	
3 	(100,100)  0.0001  10  0.1  sgd  98.43 	
4 	(200,200)  0.0001  10  0.1  sgd  98.36 	
5 	(100,100)  0.001  10  0.1  sgd  98.43 	
6 	(100,100)  0.01  10  0.1  sgd  98.43 	
7 	(100,100)  0.1  10  0.1  sgd  98.36 	
8 	(100,100)  0.001  10  0.01  sgd  98.61 	
9 	(100,100)  0.001  10  0.001  sgd  98.36 	
10	 	(100,100)  0.001  10  0.01  lbfgs  98.36 	
11	 	(100,100)  0.001  10  0.01  adam  98.36 	
 
Based  on  the  table,  the  best  parameters  for  the  MLPClassifier  are:  Hidden  layers: 
(100,100),  Alpha:  0.001,Verbose:  10,Learning_rate_init:  0.01,Solver:  'sgd'.These 
parameters were associated with the highest accuracy of 98.61%. 
 
7.4  Model Deployment 	
After  the  model  is  trained  and 	
optimized, it is deployed to classify the 
cause  of  death  from  new  cases.  To 
facilitate this process, a Graphical User 
Interface is developed to make it easy to classify the cause of death.  In addition, a module 
is  prepared  to  enable  users  to  input  the  cause  of  death  as  a  CSV  file,  and  the  model  will 
predict the  corresponding  ICD codes and  add them to a new  column labeled "code" in the 
CSV file.

145 	
Compendium of Project Reports 
 
Conclusion 
 	
Based  on  the  results  and  model  optimization,  it  can  be  concluded  that  a  machine 	
learning approach using MLPClassifier is an effective method for classification of cause of 
death.  Despite  the  unsuccessful  extraction  of  handwritten  text  from  scanned  documents, 
the results suggest that the model can be improved through fine-tuning with a large dataset 
of  handwritten  characters  commonly  used  by  doctors.  The  MLP  classifier  model  trained 
based  on  MCCD  data  set  in  excel  format  for  the  year  2022  showed  high  accuracy, 
precision,  recall,  and  F1-score  values,  indicating  that  it  correctly  classified  a  high 
percentage  of  samples.  The  best  parameters  for  the  MLPClassifier  were  determined  to  be 
Hidden layers: (100,100), Alpha: 0.001, Verbose: 10, Learning_rate_init: 0.01, and Solver: 
'sgd', with an accuracy of 98.61%. 	
 
To  facilitate  the  classification  process,  a  Graphical  User  Interface  was  developed, 	
and  a  module  was  prepared  to  enable  users  to  input  the  cause  of  death  as  a  CSV  file,  and 
the model will predict the corresponding ICD codes and add them to a new column labeled 
"code"  in  the  CSV  file.  This  can  be  a  valuable  tool  for  administrators,  policy  planners, 
researchers, and other professionals in the area of public health. 	
 
However,  it  is  important  to  note  that  the  model's  accuracy  may  be  affected  by  the 	
quality of the input data, and it may require continuous updates  and improvements to stay 
relevant and accurate. Overall, this project provides a promising approach to improving the 
accuracy and efficiency of cause of death classification. 	  	
Reference 	
i. 	Ghulam Mujtaba, Liyana Shuib, Ram Gopal Raj, Retnagowri Rajandram, Khairunisa Shaikh, 
Mohammed  Ali  Al-Garadi  :"Automated  classification  of  cause  of  death  in  verbal  autopsy 
data using machine learning" by Inoue et al. (2018). 
The paper can be found in the journal "PLoS ONE" and can be accessed at the following link: 
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0193890 	
ii. 	Organization  WHO  (1979):  Medical  Certification  of  Cause  of  Death:  instructions  to 	
physicians on use of international form of medical certification of cause of death 	
iii.	 	Department of Economics and Statistics (2020): Report on Medical Certification of cause of 	
death

146                  	Compendium  of  Project  Reports

147 	
Compendium of Project Reports 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
 	
 	Submitted  By 	
Sri. Vijay R., 	Research Officer	 	
 
 
 
 
 
 
 
 	
Exploration of land utilization in Kerala	-	 	
Forecasting	 	
 A Machine Learning Approach

148                  	Compendium  of  Project  Reports	 	
 
 
Introduction	 	
In  the  past  two  decades,  Kerala  has  experienced  significant  changes  in  land  use, 	
with  a  noticeable  decline  in  area  used  for  cultivation,  known  as  net  sown  area.  Non-
agricultural  land  use  has  gradually  increased  while  agricultural  activities  have  decreased. 
This conversion of agricultural land for non-agricultural purposes not only poses a threat to 
food  security  but  also  raises  ecological  concerns.  The  factors  that  affect  land  use  are 
complex,  including  things  like  the  surrounding  environment,  topography,  climate,  and 
policies,  etc.  Here,  we  focus  on  the  impact  of  different  types  of  land  use  on  the  area  used 
for farming. Examining the changes in land use patterns during a specific timeframe allows 
us  to  obtain  valuable  information  regarding  the  present  condition  of  agriculture  land 
utilization. 	
Predicting  time  series  data  can  be  challenging  due  to  significant  fluctuations, 	
changing trends, and limited information. Traditionally, a number of techniques, including 
univariate  Autoregressive,  univariate  Moving  Average,  Simple  Exponential  Smoothing, 
and in particular Autoregressive Integrated Moving  Average with its numerous variations, 
have been used to predict the next lag of time series data. The other most used forecasting 
methods,  multivariate  ARIMA  models  and  vector  auto-regression  models,  generalize 
univariate  ARIMA  models  and  univariate  autoregressive  models  by  allowing  several 
evolving  variables.  However,  with  the  advancements  in  computational  power  and  the 
development  of  more  advanced  machine  learning  algorithms,  such  as  deep  learning,  new 
methods  have  emerged  for  analyzing  and  forecasting  time  series  data.  Deep  learning 
approaches excel  at identifying patterns and structures in data, including non-linearity and 
complexity, in time series forecasting. 	
In  this  study,  a  deep  learning  approach  using  neural  networks  is  used  to  analyze 	
land utilization patterns in Kerala and predict the net sown area across different categories 
of land use. By analyzing numerical data on past land usage, a model is built to understand 
the  underlying  patterns  and  make  predictions  about  the  future  proportion  of  land  used  for 
cultivation. 
2. Objectives	 	
â¢ 	To identify the net sown area use changes in Kerala.  	
â¢ 	To make future prediction of net sown use area.  	
â¢ 	To understand the transformation of land.  	
â¢ 	To  restrict  land  use  that  affects  human  survival  by  implementing  effective  policies 
and regulations.  
 	
3. Methodology and method used	 	
The modeling procedure can be broken down into three simple steps as given below. 	
1.	 Data collection: Gathering the required data from various sources. 	
2.	 Data  preprocessing:  Cleaning  and  organizing  the  collected  data  to  make  it  suitable 
for analysis. 	
3.	 Model  training:  Using  the  prepared  data  to  train  the  model  and  enable  it  to  learn 
patterns and make predictions.

149 	
Compendium of Project Reports 
 
 
 	
The  Directorate  of  Economics  and  Statistics  (DES),  Government  of  Kerala,  has 	
published  reports  containing  secondary  data  on  land  use  patterns  in  Kerala.  This  data, 
spanning  from  2005-06  to  2020-21,  was  utilized  to  create  a  neural  network  model  for 
predicting  the  future  net  sown  area  in  the  state.  Various  methods  are  applied  for  data 
preprocessing to ensure that the data is in a suitable format for analysis. For this study, we 
used  the  Long  Short-Term  Memory  (LSTM)  techniques  to  build  a  model  for  the  selected 
dataset.  LSTM  is  a  special  case  of  Recurrent  Neural  Network  (RNN)  method  in  deep 
learning developed to address the challenges related to the forecasting models.  	
RNNs 	are a type of neural network that can process sequences of data. They  work 	
by taking input from the current time step and combining it with information from previous 
time  steps,  allowing  them  to  capture  temporal  dependencies  in  the  data.  However, 
traditional  RNNs  suffer  from  the  "vanishing  gradient"  problem,  where  the  gradients  that 
flow  through  the  network  become  too  small  to  be  useful  for  learning  long-term 
dependencies. 	LSTMs 	were  designed  to  overcome  the  vanishing  gradient  problem  by 	
introducing a memory cell, which can store information over time and selectively forget or 
update  it  based  on  the  input.  LSTMs  also  have  separate  gates  for  controlling  the  flow  of 
information  through  the  cell,  allowing  them  to  learn  long-term  dependencies  more 
effectively. 	
In  short,  LSTMs  are  a  powerful  type  of  RNN  that  excel  at  capturing  long-term 	
dependencies  in  data.  For  time  series  forecasting,  different  algorithms  are  available  based 
on the characteristics of the dataset. In this study, we utilized the Encoder-Decoder model, 
also  known  as  the  Autoencoder  model,  which  is  a  multi-step  LSTM  model. 	Figure  1	 	
serves as a visual representation of the sequential process followed in the study. 	
Figure 1: Work flowchart 	
 
 
 
 
 
 
 
 
 
 	
Figure 1: Process flowchart 	
 	
 
 
 
 
 
 
 
 
 	
Data	 	
Analysis and 	
Visualization	 	
Data 	
Preprocessing	 	
Training Set X 	
Training 	
Set Y	 	
LSTM Auto 	
Encoder 	
Model 	
Model Fit	 	
Predicted 	
Future 	
Optimization	 	
Evaluation

150                  	Compendium  of  Project  Reports	 	
 
4. Data Set 
4.1 Collection 	
The  secondary  data  on  land  use  area  patterns  for  the  period  from  2005-06  to     	
2020-21  was  obtained  from  the  official  website  of  the  Department  of  Economics  and 
Statistics, Kerala. The dataset covers a total geographical area of 3,886,287 hectares in the 
state.  The  land  is  classified  into  thirteen  distinct  categories,  including  Forest,  Land  put  to 
non-agricultural  use,  Barren  and  uncultivable  land,  Permanent  pastures  and  other  grazing 
land,  Land  under  miscellaneous  tree  crops,  Cultivable  waste,  Fallow  other  than  current 
fallow,  Current  fallow,  Marshy  land,  Still  water,  Waterlogged  area,  Social  forestry,  and 
Net  area  sown.  This  classification  system  enables  a  comprehensive  analysis  and 
categorization of the various land uses within the region. 
4.2 Preprocessing 	
In  order  to  develop  an  effective  model,  it  is  necessary  to  preprocess  and  transform 	
the  raw  data  into  a  suitable  format.  For  this  study,  we  have  chosen  the  LSTM  Network  as 
the  main  model  framework  due  to  its  ability  to  effectively  handle  numerical  input.  In  this 
part,  the  original  classification  of  land  use  area  into  13  categories  has  been  reconstructed 
into  8  basic  categories  based  on  the  nature  of  the  data.  Some  classifications  have 
consistently maintained negligible values throughout the entire timeframe and thus remain 
unchanged.  As  a  result  of  this  data  transformation,  a  dataset  consisting  of  144  data  points 
(16x8 samples) has been generated. The dataset is structured as follows: 	
Table 1: Land use area pattern in Kerala 	
Agriculture 	
year 	
Barren and 
uncultivable 	
land 	
Cultivable 	
waste 	
Fallow 
land 	
Forest 
area 	
Still 
water 	
Others  Non- 	
agricultural 	
use 	
Net area 
sown 	
2005-06 	26457  69165  113263  1081509  74286  19202  369922  2132483 	
2006-07 	26125  90298  128795  1081509  82702  16743  358684  2101431 	
2007-08 	25527  92764  128167  1081509  84829  12904  371558  2089029 	
2008-09 	24931  96193  113714  1081509  92876  11954  376155  2088955 	
2009-10 	22046  98014  122319  1081509  101547  10231  371906  2078715 	
2010-11 	19573  91665  127971  1081509  100111  9777  384174  2071507 	
2011-12 	17552  95437  134726  1081509  107181  9826  399924  2040132 	
2012-13 	16354  96596  132579  1081509  99789  8784  402567  2048109 	
2013-14 	13655  97069  128322  1081509  99673  9239  405826  2050994 	
2014-15 	12952  100676  120070  1081509  100453  8618  419128  2042881 	
2015-16 	13100  99489  125261  1081509  100589  8620  434646  2023073 	
2016-17 	11780  101379  127538  1081499  98343  8332  441934  2015482 	
2017-18 	10894  96491  106983  1081509  98889  8065  443041  2040415 	
2018-19 	10276  96489  102991  1081509  99323  8035  454040  2033624 	
2019-20 	10619  99810  104128  1081509  100160  8100  455897  2026064 	
2020-21 	9530  93975  97008  1081509  100033  8184  460919  2035129

151 	
Compendium of Project Reports 
 	
This dataset, derived from the reconstructed land use area classifications, will serve 	
as the basis for training and evaluating the LSTM Network model. 
5. Model Construction, Training and Optimization 	
The  LSTM  network  consists  of  input,  hidden,  and  output  layers  and  we  need  to 	
specify  parameters  such  as  the  number  of  layers  and  nodes,  learning  rate,  activation 
function,  loss  function,  and  optimizer.  To  begin,  we  construct  the  LSTM  autoencoder 
model  by  defining  its  architecture  and  parameters.  Once  the  model  is  set  up,  we  train  it 
using  a  known  training  set,  which  in  this  case  is  the  original  data  set.  During  training,  the 
model learns to encode the input data into a compressed representation and decode it back 
to reconstruct the original input. 	
During the training process, evaluation functions are employed to calculate metrics 	
such as loss and accuracy, which help assess the performance of the model. Based on these 
evaluation  results,  we  can  make  adjustments  to  the  model's  parameters  to  improve  its 
performance.  The  optimization  phase  involves  iteratively  refining  the  model  by  adjusting 
the  parameters,  such  as  the number  of layers  and  nodes,  learning  rate,  activation  function, 
loss  function,  and  optimizer,  until  we  achieve  satisfactory  results  in  terms  of  the  model's 
performance and the quality of data reconstruction. 
6. Learning Process Analysis 	
To  understand  how  the  model  learns,  we  regularly  check  its  predictions  on  the 	
training  set  at  specific  epoch  intervals.  By  comparing  these  predictions  over  time,  we  can 
analyze the changing trends and patterns in the learning process. 
7. Tools and Libraries Used 	
The  tool  used  for  building  a  Neural  Network  model  is  Google  Colab,  which  is  a 	
Python-based  platform  for  performing  various  tasks,  including  machine learning  and  deep 
learning. Colab offers a convenient and collaborative environment  where Python  code can 
be written and executed interactively. It provides a wide range of libraries and frameworks 
that are commonly used for machine learning and deep learning tasks. A brief explanation 
of  the  libraries  and  modules  used  in  the  project  for  data  handling,  model  building,  and 
visualization  are  detailed  below.	 When  creating  an  LSTM  autoencoder  model,  you  can 	
utilize  several  tools  and  libraries  to  simplify  the  implementation  process.  Here  are  some 
commonly used ones: 
Python:	  Python  is  a  popular  programming  language  for  machine  learning  and  deep 	
learning  tasks.  It  offers  a  wide  range  of  libraries  and  tools  that  can  be  used  for  building 
LSTM autoencoder models. 
NumPy: 	It's  a  library  that  helps  with  numerical  computations  and  provides  functions  to 	
work with large arrays of numbers efficiently.	 	
TensorFlow  and  Keras:	  They  are  libraries  used  for  building  and  training  machine 	
learning  models.  TensorFlow  is  the  underlying  framework,  and  Keras  is  a  user-friendly 
interface for TensorFlow. The code uses them to create neural network models.

152                  	Compendium  of  Project  Reports	 	
 
Pandas:	  It's  a  library  for  data  manipulation  and  analysis.  It  provides  data  structures  to 	
handle structured data and perform operations like filtering and merging. 
Matplotlib:	  It's  a  library  for  creating  visualizations.  It  offers  functions  to  create  different 	
types of plots like line plots and bar plots. 
Scikit-learn:	  It's a machine learning library that  provides tools for data preprocessing and 	
model evaluation. The code uses the StandardScaler class to scale or normalize the data. 
Seaborn:	  It's a library that works on top of Matplotlib and helps create visually appealing 	
statistical graphics. 
Datetime:	  It's  a  module  for  working  with  dates  and  times.  The  code  imports  the  datetime 	
class from this module. 
8. Data visualization 	
Figure 2: Graphical format of dataset

153 	
Compendium of Project Reports 
 	
  	
9. Model working rule 	
In  the  model  architecture,  the  working  rule  involves  using  the previous  three steps 	
of the data to predict the next step. The objective is to predict the net area sown for the next 
time step. The model takes the land use data from three consecutive time steps as input and 
trains  to  learn  the  patterns  and  dependencies  within  the  data.  It  then  uses  this  learned 
information to make predictions for the net area sown in the next time step. For instance, to 
predict  the  net  area  sown  for  the  year  2008-09,  the  model  would  utilize  the  land  use  data 
from  the  previous  three  years  (2005-06,  2006-07  and  2007-08).  It  would  analyze  the 
patterns  and  relationships  between  the  lands  use  categories  over  these  three  years  and 
generate a prediction for the net area sown in 2008-09. By  applying this  working  rule, the 
LSTM  autoencoder  model  aims  to  capture  the  temporal  dependencies  and  trends  in  the 
data,  allowing  for  accurate  predictions  of  the  net  area  sown  based  on  the  previous  three 
steps  of  land  use  information.  Here  is  the  table  representation  of  the  working  rule 
mentioned above. 	
Table 2: Working rule of the model 	
Training Set X  Training Set Y 	
Agricultu
re year 	
Barren 
and 	
uncultiva
ble land 	
Cultivable 	
waste 	
Fallow 
land 	
Forest 
area 	
Still 
water 	Others 	
Non-	
agricultur	
al use 	
Net area 
sown 	Net area sown 	
2005-06 	26457	 	69165	 	113263	 	1081509	 	74286	 	19202	 	369922	 	2132483	 	 	
2006-07 	26125	 	90298	 	128795	 	1081509	 	82702	 	16743	 	358684	 	2101431	 	 	
2007-08 	25527	 	92764	 	128167	 	1081509	 	84829	 	12904	 	371558	 	2089029	 	 	
2008-09 	24931	 	96193	 	113714	 	1081509	 	92876	 	11954	 	376155	 	2088955	 	2088955	 	
2009-10 	22046	 	98014	 	122319	 	1081509	 	101547	 	10231	 	371906	 	2078715	 	2078715	 	
2010-11 	19573	 	91665	 	127971	 	1081509	 	100111	 	9777	 	384174	 	2071507	 	2071507	 	
2011-12 	17552	 	95437	 	134726	 	1081509	 	107181	 	9826	 	399924	 	2040132	 	2040132	 	
2012-13 	16354	 	96596	 	132579	 	1081509	 	99789	 	8784	 	402567	 	2048109	 	2048109	 	
2013-14 	13655	 	97069	 	128322	 	1081509	 	99673	 	9239	 	405826	 	2050994	 	2050994	 	
2014-15 	12952	 	100676	 	120070	 	1081509	 	100453	 	8618	 	419128	 	2042881	 	2042881	 	
2015-16 	13100	 	99489	 	125261	 	1081509	 	100589	 	8620	 	434646	 	2023073	 	2023073	 	
2016-17 	11780	 	101379	 	127538	 	1081499	 	98343	 	8332	 	441934	 	2015482	 	2015482	 	
2017-18 	10894	 	96491	 	106983	 	1081509	 	98889	 	8065	 	443041	 	2040415	 	2040415	 	
2018-19 	10276	 	96489	 	102991	 	1081509	 	99323	 	8035	 	454040	 	2033624	 	2033624	 	
2019-20 	10619	 	99810	 	104128	 	1081509	 	100160	 	8100	 	455897	 	2026064	 	2026064	 	
2020-21 	9530	 	93975	 	97008	 	1081509	 	100033	 	8184	 	460919	 	2035129	 	2035129	 	
2021-22 	 	 	 	 	 	 	 	 	?

154                  	Compendium  of  Project  Reports	 	
 
10. Hyper Parameters Used 	
The model begins with random parameters and undergoes fine-tuning for improved 	
performance. We trained the model using the following hyperparameters: 	
ï· 	LSTM layers with 64 and 32 units respectively. 	
ï· 	Dropout regularization to prevent overfitting. 	
ï· 	Adam optimizer for adaptive learning rate. 	
ï· 	Mean squared error (MSE) loss function for training. 	
ï· 	60 epochs for training iterations. 	
ï· 	Batch size of 24 data points per batch. 	
ï· 	Divided into 13 iterations per epoch. 	
Here are simplified definitions of each hyperparameter used in the model. 	
ï§ 	LSTM layers:	 Memory cells in the model that captures long-term dependencies 	
in the data. 	
ï§ 	Dropout regularization:	 Technique to prevent overfitting by randomly ignoring 	
some input units during training. 	
ï§ 	Adam  optimizer:	  Algorithm  that  adjusts  the  learning  rate  for  each  parameter 	
during training. 	
ï§ 	Mean  squared  error  (MSE)  loss  function:	  Measures  the  average  squared 	
difference between predicted and actual values. 	
ï§ 	Epochs:	  Number  of  times  the  model  goes  through  the  entire  dataset  during 	
training. 	
ï§ 	Batch size:	 Number of data samples processed together in each training iteration. 	
ï§ 	Iterations:	 Number of times the model processes the dataset in one epoch. 	
11. Training loss and Validation loss   	
The  training  loss  assesses  how  well  the  model  reconstructs  the  training  data, 	
whereas the validation loss assesses how well it performs on new data. To achieve accurate 
reconstruction  and  good  generalization,  we  strive  to  minimize  both  losses. Monitoring  the 
validation  loss  helps  in  identifying  instances  of  overfitting,  and  modifying  the  model  or 
stopping  training  can  improve  reconstruction  quality.  The  following  graph  depicts  the 
training  and  validation  loss  during  the  training  procedure.  It  shows  how  the  loss  of  the 
model lowers over time, showing increasing performance and convergence. 	
Figure 3: Loss curve

155 	
Compendium of Project Reports 
 
12. Results with Inference 	
The  hyperparameters  used  considerably  enhanced  the  model's  training  efficiency, 	
resulting  in  accurate  and  efficient  data  analysis.  The  loss  (mse)  demonstrates  that  the 
model  learns  by  defining  a  wide  value  range  and  making  precise  modifications.  Overall, 
the  model  is  thought  to  be  valid  and  effective.  The  prediction  results  are  presented  in  the 
table below, and a visual representation is provided by the following graph. 	
Table 3: Predicted values by the model 	
Agriculture 	
year 	
Net sown area- 
Actual value 	
Net sown area- 
Predicted value 	
2005-06 	2132483	 	 	
2006-07 	2101431	 	 	
2007-08 	2089029	 	 	
2008-09 	2088955	 	2092886 	
 	2009-10 	2078715	 	2078375.2	 	
2010-11 	2071507	 	2065916.9	 	
2011-12 	2040132	 	2053325.1	 	
2012-13 	2048109	 	2046021.9	 	
2013-14 	2050994	 	2042406	 	
2014-15 	2042881	 	2034051.8	 	
2015-16 	2023073	 	2032418.8	 	
2016-17 	2015482	 	2026918.9	 	
2017-18 	2040415	 	2030204.1	 	
2018-19 	2033624	 	2032539.2	 	
2019-20 	2026064	 	2036676.2	 	
2020-21 	2035129	 	2032929.2	 	
2021-22 	 	2031854.9	 	
 	
Figure 4: Predicted Vs Actual value curve

156                  	Compendium  of  Project  Reports	 	
 	
The  dataset  used  for  the  model  is  small,  limiting  its  ability  to  learn  complicated 	
patterns.  As  a  result,  based  on  the  rules  it  has  learned,  the  model  creates  predictions.  To 
enhance  the  accuracy  of  future  predictions,  it  is  advisable  to  expand  the  dataset  by 
including more features and forecasting each feature value. The projected outcomes show a 
large decrease in the share of land used for cultivation in the agricultural year 2021-22. The 
anticipated  cultivable  land  area  for  2021-22  is  2031854.9  hectares.  This  reduction  in  land 
usage  could  potentially  have  adverse  effects  on  food  production,  raising  concerns  about 
meeting the growing demand. 
Conclusion	 	
This  study  emphasizes  the  advantages  of  forecasting  land  use  data  using  deep 	
learning  techniques.  The  algorithm  can  more  accurately  forecast  the  net  sown  area  based 
on  other  land  uses  by  examining  a  broader  and  more  diverse  dataset.  Although  the  model 
has  revealed  some  patterns,  it's  critical  to  keep  in  mind  that  real-world  situations  are 
intricate and dynamic. By experimenting with various modifications, we may optimize the 
hyperparameters to improve the performance of the model. Altering the  model parameters 
and  adding  more  affecting  variables  to  the  dataset  are  additional  steps  in  enhancing  the 
findings.  The  study  was  able  to  effectively  forecast  the  outcomes  to  a  significant  extent. 
However, the  remaining  goals  were temporarily put on hold because there weren't enough 
feature values available. 	
 	
Reference: 	
1.    Imad Basheer and M. Hajmeer: 2001: Artificial Neural Networks- fundamentals, computing, 	
design and application.  	
2.    Sima  Siami-Namini, Neda  Tavakoli,  Akbar Siami  Namin:  2018-  A  Comparison  of  ARIMA  and 	
LSTM  in  Forecasting  Time  Series-  17th  IEEE  International  Conference  on  Machine  Learning 
and Applications. 
Various  deep  learning  architectures,  such  as  convolutional  neural  networks  (CNNs), 
recurrent  neural  networks  (RNNs),  and  generative  adversarial  networks  (GANs),  have  been 
applied  on  various  numerical  historical  data  and  demonstrated  the  superiority  of  deep 
learning  methods  over  traditional  machine  learning  methods  in  terms  of  accuracy  in  these 
papers.

157 	
Compendium of Project Reports 
 
  	
Building Construction Cost and Index 	
Prediction in Kerala 	 	
A Machine Learning Approach	
 	
Submitted  By 	
Smt. Dhanya A., 	Deputy Director

158                  	Compendium  of  Project  Reports	 	
 
INTRODUCTION  	
The  construction  industry  is  a  vital  sector  in  any  economy,  and  construction  cost 	
estimation  is  an  essential  aspect  of  the  industry.  In  Kerala,  the  building  construction  cost 
index  (BCCI)  and  the  construction  cost  are  significant  determinants  of  the  viability  of 
construction  projects.  Building  construction  is  a  complex  and  multi-disciplinary  process 
that  requires  the  integration  of  various  factors  such  as  labor,  materials,  and  time.  The cost 
of  construction  is  an  important  aspect  of  the  process,  as  it  affects  the  feasibility  of  the 
project  and  the  return  on  investment  for  the  stakeholders.  In  recent  years,  the  building 
construction industry in Kerala has seen significant growth, with an increasing demand for 
new  construction  projects.  However,  the  estimation  of  construction  costs  can  be 
challenging,  as  it  is  influenced  by  many  variables  such  as  inflation,  changes  in  market 
conditions,  and  advances  in  construction  technologies.  The  traditional  methods  of 
estimating construction costs are often time-consuming, inaccurate, and subjective.  	
In  order  to  address  these  challenges,  the  field  of  construction  cost  prediction  has 	
seen  significant  progress  in  recent  years.  The  integration  of  machine  learning  (ML) 
techniques  in  construction  cost  estimation  can  provide  a  more  accurate  and  efficient 
solution.    Machine  learning  techniques  have  been  used  to  develop  cost  prediction  models 
that  can  provide  more  accurate  and  reliable  estimates  of  construction  costs.  The  objective 
of this project is to utilize these advanced techniques to develop a machine learning model 
that  can  be  used  to  forecast  the  Building  Construction  Cost  Index  (BCCI)  and  the  cost  of 
building construction in Kerala. The project also aims to develop a graphical user interface 
(GUI)  that  can  be  used  by  stakeholders  to  predict  the  cost  of  construction  for  future 
projects.  By  providing  an  accurate  and  reliable  tool  for  construction  cost  prediction,  this 
project  aims  to  support  the  growth  of  the  construction  industry  in  Kerala  and  provide 
valuable insights for stakeholders in the industry. 	
   The results of this project will  provide valuable insights into the potential of ML 	
in construction cost estimation and will benefit the construction industry in Kerala. 	
 	
OBJECTIVE  	
The 	objective of this study is to utilize Machine Learning techniques to predict the 	
future  Building  Construction  Cost  Index  (BCCI)  and  building  construction  cost  in  Kerala. 
This  will  help  in  providing  better  cost  estimates  for  builders  and  contractors,  thus 
improving  the  efficiency  of  the  construction  industry.  The  study  will  also  develop  a 
graphical user interface (GUI) for ease of use by construction professionals. The results of 
this  study  will  be  useful  for future  reference  and  as  a  starting  point  for  further  research  in 
this field. 
 
METHODOLOGY  	
i. 	Data 	Collection:  	
Building  material  price  and  Labour  wage  rates  from  2011-12  to  2021-22  were 	
collected from Department of Economics and Statistics.  	
ii.	 	Data Preparation: 
Districts  in  Kerala  were  given  codes  from  Thiruvananthapuram  to  Kasaragod  as  1 	
to  14  respectivey.  Then  computed  construction  cost  from  building  material  price,  labour 
charges  and  other  charges  collected.  After  that  prepared  the  weighing  diagram  and  the 
construction cost index 	for all districts.

159 	
Compendium of Project Reports 
 	
In  the  building 	construction  cost  index  data,  the  variables  are  year,  district  code 	
and BCCI. There are no null values in this dataset. 	
In  the  building  construction  cost  data,  the  variables  are  year,  district  code,  area  in 	
square  feet,  material  costs,  labour  costs,  other  charges  and  the  total  construction  cost. 
There are no null values in this dataset. 	
iii.	 	Model Selection: 
In  this  study,  the  goal  was  to  predict  building  construction  costs  and  cost  indices 	
using  machine  learning  techniques.  After  exploring  various  models,  multiple  linear 
regression was selected as the most appropriate model for this problem. 	
Multiple linear regression is a statistical method that models the linear relationship 	
between  a  dependent  variable  (building  construction  cost  or  cost  index)  and  independent 
variables (features such as location, materials cost, labour costs, etc.). The model estimates 
the coefficients for each feature, which can then be used to make predictions based on new 
data. 	
The equation for multiple linear regression can be represented as: 
y = Î²	0 + Î²	1X	1 + Î²	2X	2 + ... + Î²	nX	n 	
where: 
y is the dependent variable (the one being predicted) 
X	1, X	2, ..., X	n are the independent variables (the predictors) 	
Î²0 is the intercept term (the value of y when all X's are 0) 	
Î²1,  Î²	2,  ...,  Î²n  are  the  coefficients,  representing  the  strength  and  direction  of  the 	
relationship between each independent variable and the dependent variable. 	
The goal of multiple linear regression is to estimate the values of the coefficients Î²	0, 	
Î²1, Î²	2, ..., Î²	n that best fit the observed data. These estimates can be obtained using various 	
techniques, such as ordinary least squares or maximum likelihood estimation. 	
The choice of multiple linear regression was based on several factors, including: 	
ï· 	Simplicity:  Multiple  linear  regression  is  a  simple  and  well-understood 
model that can be easily interpreted and applied to this problem. 	
ï· 	Linearity:  The  relationship  between  the  dependent  variable  and  the 
independent  variables  was  assumed  to  be  linear,  which  is  a  reasonable 
assumption for this problem. 	
ï· 	Availability  of  data:  Adequate  data  was  available  to  train  the  model  and 
validate its performance. 	
ï· 	Performance: In initial experiments, multiple linear regression was found to 
perform  well  in  terms  of  accuracy  and  stability  compared  to  other  models 
such as decision trees, random forests or ANN. 	
In conclusion, multiple linear regression was selected as the model for prediction in 	
this  study,  as it offered  a good balance between  performance and interpretability, and was 
well-suited to the problem and available data.   	
iv.	 	Model Training and Validation: 
The  multiple  linear  regression  model  was  trained  on  a  dataset  of  building 	
construction  costs  and  cost  indices.  The  dataset  was  divided  into  two  parts:  80%  for 
training and 20% for validation.

160                  	Compendium  of  Project  Reports	 	
 	
The  training  process  involved  estimating  the  coefficients  for  each  feature  in  the 	
model  using  the  training  data.  This  was  done  using  the  ordinary  least  squares  method, 
which  minimizes  the  sum  of  the  squared  residuals  between  the  observed  and  predicted 
values. 	
Once  the  model  was  trained,  its  performance  was  evaluated  on  the  validation  data 	
using two metrics: mean squared error (MSE) and coefficient of determination (R-squared). 
The  MSE  measures  the  average  difference  between  the  observed  and  predicted  values, 
while the R-squared measures the proportion of variance in the dependent variable that can 
be explained by the independent variables. 	 
a.	 	Building Construction Cost Index (BCCI):  	
The  results  showed  that  the multiple  linear  regression  model  had  a MSE  of 	
0.2667  and  an  R-squared  of  0.733  on the validation  data,  indicating  that  the model 
had good performance in terms of both accuracy and goodness-of-fit. These results 
confirmed that multiple linear regression was a suitable model for this problem and 
provided a good basis for making predictions on new data. 	
In  conclusion,  the  multiple  linear  regression  model  was  trained  and 	
validated  on  the  data,  and  its  performance  was  evaluated  using  two  appropriate 
metrics.  The  results  showed  that  the  model  had  good  performance  in  terms  of 
accuracy and goodness-of-fit, and was ready for testing on new data.

161 	
Compendium of Project Reports 
 	
b.	 	Building Construction Cost:  	
The model was trained on 80% of the data and tested on the remaining 20%. 	
The  mean  squared  error  (MSE)  and  R-squared  metrics  were  used  to  evaluate  the 
model's performance on the testing set. 	
The  mean  squared  error  (MSE) on  the testing  set  was  found  to  be  0.00688, 	
and the R-squared was 0.918. These values indicate that the model has a good fit to 
the data, with a small difference between the actual and predicted values (MSE) and 
a high proportion(91.8%) of the variance in the dependent variable explained by the 
independent variables in the model (R-squared). This is a high value, indicating that 
the model has a good fit to the data. 
 	
 	
 	
 	
v.	 	Model deployment:  
The  result  of  2938154.66  that  got  from  testing  the  model  with  new  values 	
such  as  year  -  2022,  district  code  -  14  (Kasaragod),  and  area  in  square  feet  -  2000 
indicates that the model has made a prediction for the dependent variable based on 
the  given  independent  variables.  This  prediction  is  a  good  one,  suggests  that  the 
model is likely to make accurate predictions in similar scenarios.

162                  	Compendium  of  Project  Reports	 	
 	
Also developed a Graphical User Interface(GUI) for predicting the Building 	
Construction Cost	 using âtkinterâ package and developed a mobile application using 	
âkivyâ package in python. 	
 	
 	
 	
 	
 
CONCLUSION: 
 	
The  results  of  the  model  testing  demonstrate  that  the  developed  machine  learning 	
model has good performance in terms of both accuracy and fit to the data.  	
Here  also  developed  a  graphical  user  interface  (GUI)  and  mobile  application  for 	
ease  of  use  by  construction  professionals.  The  results  of  this  study  will  be  useful  for 
providing  better  cost  estimates  for  builders  and  contractors,  thus  improving  the  efficiency 
of the construction industry. 	
Overall,  the  results  of  this  study  fulfil  the  objective  of  utilizing  machine  learning 	
techniques  to  predict  the  future  BCCI  and  building  construction  cost  in  Kerala.  The 
model's  performance  and  the  developed  GUI  provide  a  starting  point  for  future  reference 
and further research in this field.  	
Future  work  could  include  expanding  the  dataset  to  include  additional  factors  that 	
may  influence  construction  costs,  such  as  the  size  and  complexity  of  the  building  or  the 
economic  conditions  at  the  time  of  construction.  Additionally,  further  research  could  be 
conducted  to  compare  the  performance  of  different  machine  learning  models  and  to 
evaluate the long-term effectiveness of the tool in predicting construction costs. 	
Predictions  like  this  can  be  made  and  published  on  the  website  of  Department  of 	
Economics and Statistics in future.

163 	
Compendium of Project Reports 
 	
 	
REFERENCE: 
 	
ï· 	Website of  	
1.	 	Department of Economics and Statistics, Kerala 	
2.	 	Department of Public Works Kerala 	
3.	 	Kerala State Housing Board 	
4.	 	Central Public Works Department 	
ï· 	International  Journal  of  Civil  Engineering  and  Technology  (IJCIET),  Construction  costs  in  affordable 
housing  in  kerala:  relative  significance  of  the  various  elements  of  costs  of  affordable  housing 
projects, Manoj P K. 	
ï· 	International Journal of Earth Sciences and Engineering, Study on factors Affecting the performance 
of  construction  projects  and  developing a cost prediction model  using  ANN,  Melba Alias, Dhanya R 
and Gangapathy Ramaswamy, October, 2015.

164                  	Compendium  of  Project  Reports

165 	
Compendium of Project Reports 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
 
 
 
 
 
  
 	
To predict daily prices of essential 	
commodities	
 	
Submitted  By 	
Sri. Abhilash K.V., 	Research Officer

166                  	Compendium  of  Project  Reports	 	
 
 
Introduction 
 	
Predicting  the  daily  price  of  an  essential  commodity  can  be  a  valuable  tool  for 	
businesses,  investors,  and  governments.  The  price  of  essential  commodities  such  as  oil, 
rice,  pulses  can  have  a  significant  impact  on  the  economy  and  can  be  influenced  by  a 
variety of factors such as supply and demand, economic conditions, and geopolitical events. 	
Accurate  predictions  of  the  daily  price  of  an  essential  commodity  can  help 	
businesses make informed decisions about production, pricing, and purchasing.  It  can also 
help  investors  make  informed  decisions  about  buying  and  selling  stocks  or  commodities. 
Governments  may  also  use  price  predictions  to  make  policy  decisions  and  to  manage  the 
impact of commodity price changes on the economy. 	
There are a number of statistical methods that can be used to predict the daily price 	
of  an  essential  commodity,  including  the  LSTM  Model  (Long  Short  -Term  Memory).  In 
this  approach,  a  statistical  model  is  built  using  data  on  the  prices  of  the  commodity  over 
time  and  a  number  of  independent  variables  that  are  believed  to  influence  the  price.  The 
model is then used to make predictions about the price of the commodity in the future. 	
 	
Objectives 	
â To create a model that will be able to analyze the relationship between district wise 
monthly average price data of some important essential commodities. 	
â The level of accuracy of this model may help the government to make decisions to 
ensure  the  buffer  stocks  for  commodities  in  the  market  and  minimize  the  artificial 
shortages. 	
â To prepare charts and animated maps for the better visualization of data. 	
 	
Literature Survey 
 	There has been a significant amount of research on predicting the monthly average 	
price of essential commodities. Here are a few examples of studies that have used 
regression analysis to predict commodity prices: 	
â 	Predicting Commodity Prices Using a Hybrid Neural Network Model" (Wang et al., 
2013) - In this study, the authors develop a hybrid neural network model to predict 
the  prices  of  four  commodities  (crude  oil,  natural  gas,  corn,  and  soybeans)  over  a 
five-year  period.  They  find  that  the  model  is  able  to  make  accurate  short-term  and 
medium-term predictions for all four commodities. 	
â 	Commodity Price Forecasting: A Review of the Literature" (Bilal and Sun, 2018) - 
This  review  paper  surveys  the  literature  on  commodity  price  forecasting  and 
discusses the various  statistical and machine learning methods that have been used 
to  predict  prices.  The  authors  also  discuss  the  challenges  and  limitations  of 
commodity price forecasting and suggest directions for future research. 
 	
Proposed Methodology  	
To  predict  the  daily  price  of  an  essential  commodity,  use  a  LSTM  (Long  Short-	
Term Memory) model.  	
LSTM  is  a  type  of  Recurrent  Neural  Network  (RNN)  architecture  specifically 	
designed  to  handle  sequential  data,  such  as  time  series,  natural  language  processing,  and 
speech recognition.

167 	
Compendium of Project Reports 
 	
To  build  a  regression  model  to  predict  the  daily  average  price  of  an  essential 	
commodity,  collect  data  on  the  prices  of  the  commodity  over  time.  This  collected  data 
includes the month, the year, and the price of the commodity for each month. 	 	
Analysis 
 	
We  will  use  the  following  techniques  to  analyze  the  data  and  make  the  price 	
predictions: 	
â 	Time series analysis to identify trends and patterns in the historical price 
data 	
â 	Using LSTM Model ( Long Short -Term Memory ) to develop the statistical 
model for forecasting future prices 
 	
Daily Price Data ( Jan 2015- Jan2022 ) 	
 	
distid  itemcode  mnth_year  avg_price 
01  A001  01-2015  30.00 
01  A001  01-2016  30.17 
01  A001  01-2017  30.96 
01  A001  01-2018  35.00 
01  A001  01-2019  34.88 
01  A001  01-2020  34.20 
01  A001  01-2021  33.00 
02  A001  02-2015  29.91 
02  A001  02-2016  30.61 
02  A001  02-2017  35.41 
02  A001  02-2018  35.00 
02  A001  02-2019  34.50 
02  A001  02-2020  33.50 
02  A001  02-2021  31.57 
03  A001  02-2022  37.00 	
 	
Districts Name and Codes 	
 	
Distid  District_name 	
01  Kasargod 
03  Wayanad 
04  Kozhikkode 
06  Palakkad 
08  Ernakulam 
14  Thiruvananthapuram 	
Item Names and Codes	 	
itemcode  item_name 
A001  Rice(red) - Matta 
A008  Rice(white)- Andhra Vella 
B006  Tur Dhall 
D006  Vanaspathi  (Dalda) 
E002  Chillies dry 
F003  Potato 
G001  Onion Big 
G014  Tomato

168                  	Compendium  of  Project  Reports	 	
 
Scatter Plots 	
 	
Line Plots  	
 	
LSTM (Long Short-Term Memory) 
 	
data.shape - (85595,4) 	
 
 
 
 	 	
Data columns (total 4 columns):	 	
#   Column     Non-Null Count  Dtype	 	
---  ------     --------------  -----	 	
0   distid     89595 non-null  int64	 	
1   itemid     89595 non-null  object	 	
2   mnth_year  89595 non-null  object	 	
3   avg_price  89595 non-null  float64	 	
# G014-Tomato 	
options = ['G014'] 	
# selecting rows based on condition 	
data = data.loc[ (data['distid'] == 1)  & 	
data['itemid'].isin(options)] 	
df=data[['mnth_year','avg_price']] 	
df.shape 	
(1846, 2)

169 	
Compendium of Project Reports 
 
 
In this model with a batch size of 512 and 400 epochs, the mean absolute error achieved is 
1.771214348561055. 	
ï· 	In an LSTM model, batch size and epochs are two important hyper parameters that 
need to be set when training the model. 	
ï· 	Batch  size  refers  to  the  number  of  samples  that  are  processed  in  one 
forward/backward pass.  It is a hyper parameter that determines how many samples 
are processed at a time before the weights of the model are updated. A larger batch 
size  will  reduce  the  number  of  weight  updates,  but  also  increases  the  memory 
requirements and can slow down the training process. 	
ï· 	Epochs refer to the number of times the entire training dataset is passed through the 
model.  One  epoch  means  that  the  model  has  seen  all  the  samples  in  the  training 
dataset once.  In each  epoch, the model uses the training data to update its  weights. 
Increasing the number of epochs allows the model to see more of the training data, 
which can improve the model's performance, but also increases the training time. 	
The mean absolute error can be used to evaluate the performance of an LSTM model. 
The  mean  absolute  error  (MAE)  is  a  measure  of  the  difference  between  the  predicted 
values  and  the  true  values  in  a  regression  problem.  In  the  case  of  an  LSTM  model  with  a 
mean  absolute  error  of  1.771214348561055,  it  indicates  that  on  average,  the  predictions 
made  by  the  model  are  off  by  approximately  1.77  units  (the  units  depend  on  the  scale  of 
the  target  variable).  A  lower  mean  absolute  error  indicates  that  the model  is  making  more 
accurate predictions.	 	
 
Actual - Predicted (LSTM)

170                  	Compendium  of  Project  Reports	 	
 	
 
 
 	
Future Prediction 
Look Back - 90 days 	
No . of Future Predictions - 45 days 	
 	
[[28.        ] 
[32.36255264] 
[36.07043266] 
[39.60783195] 
[43.12387085] 
[46.62206268] 
[50.12095261] 
[53.60870361] 
[56.889431  ] 
[59.77647781] 
[62.09675598] 
[63.73423004] 
[64.64855194] 
[64.85329437] 
[64.36707687] 
[63.30108643] 
[61.87174606] 
[60.35391617] 
[59.0037384 ] 
[58.00448227] 
[57.43686295] 
[57.25453186] 
[57.31728363]

171 	
Compendium of Project Reports 
 	
The  "look  back"  in  an  LSTM  model  refers  to  the  number  of  time  steps  that  the 	
model  uses  as  input  to  make  a  prediction.  The  look  back  is  an  important  hyper  parameter 
that needs to be set when training an LSTM model 
Expected Outcome 	
The expected outcome of predicting the daily price of an essential commodity is to 	
have  an  estimate  of  the  expected  price  of  the  commodity  for  the  coming  month.  This 
prediction can be used for a variety of purposes, such as: 	
Planning and budgeting 	: A consumer or a business that relies on the commodity, 	
knowing the expected price can help you plan and budget for your expenses. 	
Investment 	: An investor, knowing the expected price can help you make informed 	
decisions about whether to buy or sell the commodity. 	
Overall,  the  expected  outcome  of  predicting  the  monthly  average  price  of  an 	
essential  commodity  is  to  have  a  better  understanding  of  the  market  and  make  more 
informed decisions. 	
 	
Conclusion 
 	
General  conclusion  about  predicting  the  monthly  average  price  of  an  essential 	
commodity  is  a tedious  task,  as  the  specific  commodity  and  market  conditions  will  play  a 
large role in the accuracy and usefulness of any predictions. However, some general points 
that could be made include: 	
â 	Accurate  price  predictions  can  be  very  valuable,  as  they  can  help  Government, 
businesses and investors make informed decisions and manage risks. However, it is 
important to remember that no prediction is ever 100% certain, and there is always 
a degree of uncertainty and risk involved. 	
â 	It  is  important  to  regularly  review  and  update  predictions  as  new  information 
becomes  available  and  market  conditions  change.  This  can  help  ensure  that  your 
predictions remain relevant and accurate.

172                  	Compendium  of  Project  Reports

173 	
Compendium of Project Reports 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  	
Submitted  By 	
Smt. Maya R., 	Deputy Director	 	
 
 
 
 
 
 
 
 
 	
Forecasting of Consumer Price Index in 	
Kerala 	 	
A Machine Learning approach

174                  	Compendium  of  Project  Reports	 	
 
Abstract 
 	
This  research  paper  presents  a  comparative  analysis  of  two  popular  machine 
learning models, namely the Auto Regressive Integrated Moving Average (ARIMA) 
and  Long  Short-Term  Memory  (LSTM),  for  the  purpose  of  forecasting  Consumer 
Price  Index  (CPI)  values.  The  study  uses  historical  CPI  data  from  a  particular 
region  and  employs  both  ARIMA  and  LSTM  models  to  predict  the  CPI  values  for 
upcoming  years.  The  accuracy  of  the  models  is  measured  using  the  Mean  Squared 
Error  (MSE)  metric,  and  the  results  show  that  the  LSTM  model  is  more  efficient 
than the ARIMA model, with an MSE of 1.77 compared to 3.27 for SARIMA. The 
study  thus  provides  a  valuable  comparison  of  these  two  popular  models  and 
highlights  the  potential  benefits  of  using  LSTM  in  forecasting  time  series  data  for 
CPI values. 	
   
INTRODUCTION 
 
  The  Consumer  Price  Index  (CPI)  is  a  measure  of  the  average  change  in  prices  of 
goods  and  services  consumed  by  households  over  time.  It  is  used  to  track  changes  in  the 
cost of living and is calculated by comparing the price of a basket of goods and services in 
a  base  period  to  the  current  period.  The  basket  of  goods  and  services  is  designed  to 
represent  what  households  typically  purchase.  The  CPI  is  used  for  various  purposes 
including adjusting wages, pensions, DA and other payments to account for changes in the 
cost  of  living.  It  is  also  used  to  adjust  the  value  of  money  in  financial  contracts  such  as 
mortgages  and  bonds.  Central  banks  often  use  the  CPI  as  a  benchmark  for  inflation 
targeting and also used for adjusting tax brackets and other economic variables. 
  CPI  is  an  index  of  prices  of  commodities  in  a  fixed  quantity  over a  fixed  period  It 
typically  includes  essential  items  such  as  food,  education,  medicine,  housing,  fuel,  and 
other  necessities.  The  prices  of  these  goods  and  services  are  measured  in  relation  to  a 
reference period, known as the base period, where the value of the index is set at 100. The 
rate  at  which  the  CPI  increases  or  decreases  indicates  the  overall  change  in  the  level  of 
prices.  The  CPI  is  the  most  widely  used  measure  of  inflation,  closely  followed  by 
policymakers, financial markets, businesses, and consumers. Using predicted CPI, inflation 
rate  can  be  calculated  for  the  future.  Forecasting  the  Consumer  Price  Index  (CPI)  is  an 
important task as it provides insight into the future movements of inflation. By forecasting 
the  CPI,  policymakers,  businesses,  and  individuals  can  make  informed  decisions  and  plan 
for  the  future.  For  example,  central  banks  use  the  CPI  to  inform  their  monetary  policy 
decisions  and  businesses  use  it  to  set  prices  and  make  investment  decisions.  Additionally, 
forecasting  the  CPI  can  also  help  in  identifying  potential  economic  trends  and  detecting 
any signs of economic instability.  In general, the ability to forecast the CPI is essential for 
any  economic  decision-making,  as  it  provides  a  clear picture  of  the direction  in  which  the 
economy is heading and allows for appropriate actions to be taken. the state government of 
Kerala uses the CPI to set the minimum wages for workers and to adjust the subsidies and 
other welfare schemes.  Businesses in Kerala, particularly those in the retail and wholesale 
trade sectors, use the CPI to make pricing and investment decisions. 
 
  The CPI (IW)(CPI  Industrial workers and Agricultural  Labourers and CPI  (R\U\C) 
(CPI Rural/Urban /Combined) are released by the Department of Economics and Statistics

175 	
Compendium of Project Reports 
 
on  a  monthly  basis  and  are  widely  used  as  an  economic  indicator.  The  Consumer  Price 
Index  Numbers  for  Agricultural  and  Industrial  workers  with  base  2011-12  in  the  state  of 
Kerala  are  being  computed  monthly  for  seventeen  centers.  Apart  from  this  CPI  (R\U\C) 
with  base  2018  is  also  calculated  for  rural  and  urban  areas  of  all  districts  and  the  state 
monthly. 
  In  recent  years,  machine  learning  techniques  have  been  used  to  forecast  the  CPI. 
These techniques can take into account  a wide range of variables and patterns in historical 
data and can produce more accurate forecasts than traditional methods. 
  To  the  best  of  my  knowledge,  there  is  currently  no  study  that  has  been  conducted 
on  the  use  of machine learning  for  forecasting  the  Consumer Price  Index  (CPI)  in  Kerala. 
However, it is possible that machine learning techniques could be applied in this context to 
improve the accuracy of CPI forecasting in the state. 
  This  study  aims  to  identify  an  appropriate  forecasting  model  for  the  Consumer 
Price  Index  (CPI)  in  the  Indian  state  of  Kerala.  It  will  use  various  statistical  and  machine 
learning techniques to forecast the CPI and evaluate the performance of different models to 
determine the most suitable one. 
 
OBJECTIVE 
 
To implement machine learning models to forecast the consumer price index Kerala.  
 
LITERATURE SURVEY 
 
  There  have  been  various  studies  on  using  machine  learning  techniques  for 
forecasting consumer price indices in  India  and other countries and regions. These studies 
have  employed  techniques  such  as  time  series  analysis  and  various  machine  learning 
algorithms, such as linear regression, decision trees, and neural networks, Long Short Term 
Memory,  to  analyse  historical  data  on  consumer  prices  in  order  to  make predictions  about 
future price movements. 
 
[1]  Dr  P  K  Sarangi,  Dr  Deepthi  Sinha,  Dr  Sachin  Sinha,  Dr  Meenakshi  Sharma  in  their 
study  âForecasting  of  Consumer  Price  Index  Using  Neural  Networks  Modelâ  suggested 
ANN  as  abetter  choice  for  predicting  CPI  taking  into  account  two  network  architecture. 
The  first  one  is  12-12-1,  and  the  other  one  is  8-8-1.  The  MAE  and  MSE  for  12-12-1  is 
0.40446  and  0.28256  respectively.  While  the  MAE  and  MSE  for  8-8-1  is  0.3998  and 
0.30943 respectively. 
 
[2]  The  study  "Machine  Learning  Approach  for  the  Prediction  of  Consumer  Food  Price 
Index"  by  P.  K.  Sarangi,  D.  Sinha,  and  Mittal  (2021)  concluded  that  a  simple  Artificial 
Neural  Network  (ANN)  model  with  backpropagation  is  highly  capable  in  forecasting 
Consumer Food Price Index (CFPI).  
 
[3]  S  Zahara  et  al  2020  conducted  a  study  on  Consumer  Price  Index  Numbers  and 
suggested  a  forecasting  model,  Long  Short  Term  Memory  with  nonlinear  parameter  input 
to predict the next move in Indonesia. They used various optimization algorithms to obtain 
best  accuracy  in  the  LSTM  method.  Stochastic  Gradient  Descent,  Root  Mean  Square 
Propagation,  Adaptive  Gradient,  Adaptive  moment,  Adadelta,  Nesterov  Adam,Adamax 
optimization  algorithms  were  used  to  improve  the  accuracy  ,Nesterov  Adam  got  the  best 
result with RMSE value 4.088.But they conclude that  the accuracy in this model was still

176                  	Compendium  of  Project  Reports	 	
 
far from expectations ,there were several aspect of evaluation could be implemented in the 
next study like variation of epoch ,hidden layer, batch size input variable could be tested. 
 
[4]  Cheng  Yang  and  Shuhua  Guo  2021  in  their  research  article  âInflation  Prediction 
Method  Based  on  Deep  Learningâ  presents  a  method  for  forecasting  inflation  using  a 
GRU-RNN  (gated  RNN)  model  which  shows  superior  performance  in  predicting  the 
Consumer  Price  Index  (CPI)  compared  to  several  comparison  methods,  as  evidenced  by 
lower  MSE,  MAPE,  and  SMAPE  values.  GRU-RNN  is  an  improved  version  of  RNN 
(recurrent  neural  network)  and  has  been  shown  to  have  better  performance  in  sequence 
prediction  performance.  Furthermore,  the  GRU-RNN  model  outperforms  the  BP  method, 
which  is  a  shallow-layer  neural  network,  highlighting  the  advantages  of  using  a  deep 
network  structure  and  the  improved  capabilities  of  the  GRU-RNN  model,  which  includes 
gate nodes to enhance its performance. In this method, Consumer Price Index (CPI) is used 
as the indicator for inflation and multiple economic-related indexes are included as features 
for  the  model.  The  proposed  method  uses  historical  data  as  input  and  trains  a  GRU-RNN 
model to optimize its parameters. 
  
DATA SET AND ANALYSIS    
  The dataset used in this study consists of monthly Consumer Price Index (CPI) data 
for the state of Kerala, India. The data spans from April 2015 to September 2022, covering 
a  period  of  over  seven  years.  By  analyzing  this  dataset,  the  study  aims  to  identify  trends 
and  patterns  in  CPI  values  over  time  and  to  develop  accurate  forecasting  models  using 
machine learning techniques such as ARIMA and LSTM. 
  	
Month  CPI 
2015-01  133 
2015-02  133 
2015-03  133 
^^^^^^^^^^^^^^^^^^^^ 
2022-05  193 
2022-06  195 
2022-07  195 
2022-08  195 
2022-09  194 	
 
  	
 	
Seasonal decomposition graph 	
 	
130
150
170
190	
01-01-2015
01-08-2015
01-03-2016
01-10-2016
01-05-2017
01-12-2017
01-07-2018
01-02-2019
01-09-2019
01-04-2020
01-11-2020
01-06-2021
01-01-2022
01-08-2022	
Monthly CPI Kerala-2015  Jan to 	
2022 Sept

177 	
Compendium of Project Reports 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
METHODOLOGY AND METHODS USED  
  	
Two models are selected (ARIMA and  LSTM) and compare these models in terms 	
of  their  ability  to  analyze  and  make  predictions  based  on  the  given  dataset.The  following 
steps are implemented: 
â  Collect a dataset containing time series data on the Consumer Price Index  
â  Pre-process the data by cleaning, formatting, and potentially removing any irrelevant 	
or redundant information. 	
â  Divide the data into training and test sets or use cross-validation methods to evaluate 	
the model's performance. 	
â  Train the chosen model using the training data and fine-tune the model's parameters 	
to improve its performance. 	
â  Evaluate the model's performance using the test set. 
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 	
ANALYSIS& 	
VISUALIZATION OF 	
DATA	 	
TRAINING 	
S	E	T X	 	
DATA 	
PROCESSING 	
LSTM 	
SARIMA	 	
DATA 	
MODEL FIT 	
TESTING SET 	
PREDICTING

178                  	Compendium  of  Project  Reports	 	
 
Libraries and Tools   
  The  analysis  was  done  using  Python  Programming.    The  technique  used  for  time 
series analysis was SARIMA and LSTM.  
Libraries are, 
 	
â¢ 	pandas  	
â¢  matplotlib.pyplot  
â¢  numpy  
â¢  tensorflow  
â¢  tensorflow.keras.models  
â¢  tensorflow.keras.layers  
â¢  sklearn.preprocessing  
â¢  sklearn.metrics  
â¢  seaborn  
â¢  statsmodels.tools.eval_measures  
â¢  matplotlib.pylab  
â¢  statsmodels.tsa.seasonal 
â¢  pmdarima.arima 	
 
RESULT WITH INFERENCE   
ARIMA MODEL  
Train -Test Split 
  In  machine  learning,  it  is  important  to  divide  the  available  data  into  two  separate 
sets: training data and testing data. The purpose of this division is to use the training data to 
develop  a  predictive  model  that  can  accurately  capture  the  underlying  patterns  and  trends 
in  the  data.  Once  the  model  is  fitted  to  the  training  data,  it  can  be  evaluated  and  tested 
using the testing data to measure its effectiveness in predicting unseen data.  
  For the specific case of the ARIMA (Seasonal Auto Regressive  Integrated Moving 
Average)  model,  the  first  step  would  be  to  split  the  available  CPI  data  into  training  and 
testing datasets. The training dataset would be used to fit the SARIMA model and tune its 
parameters,  while  the  testing  dataset  would  be  used  to  evaluate  the  performance  of  the 
model in predicting future CPI values. 	
 	
 	
Training data in blue colour and test shown in orange colour

179 	
Compendium of Project Reports 
 
  In time series  analysis, a series is considered to be stationary if its mean,  variance, 
and autocorrelation remain constant over time. A stationary time series is characterized by 
a  correlogram  or  ACF  (Auto  Correlation  Function)  that  quickly  dies  down,  while  a  non-
stationary  series exhibits a slow decay. To determine stationarity, the Augmented Dickey-
Fuller (ADF) test is commonly used. The test has two hypotheses: H0, the null hypothesis, 
and  H1,  the alternative  hypothesis.  H0  assumes  that  the  series  is  non-stationary,  while  H1 
suggests that the series is stationary. If the p-value is greater than 0.05, we fail to reject H0, 
indicating  that  the  series  is  non-stationary.  If  the  p-value  is  less  than  0.05,  we  accept  H1, 
implying  that  the  series  is  stationary.Here  ADFuller  test  gives  Test  Statistic          0.575829 
and p-value .987004 that implies this data set is non stationary. 
 
Before   
  The  method  used  for  converting  non-stationary  to  stationary  for  time  series  is 
differencing.the below graph shows ACF and PACF  before and after of differencing. The 
model  identification  is  done  using  Auto  Correlation  Function  (ACF)  and  Partial  Auto 
Correlation Function (PACF). 
 
After  
 	
 	
 	
 	
 	
Graph of auto correlation and partial correlation before and after differencing

180                  	Compendium  of  Project  Reports	 	
 
  The pmdarima.arima module is a machine learning library used to identify accurate 
ARIMA  models  for  time  series  forecasting.  The  pmdarima.arima  module  provides  a 
number  of  functions  that  simplify  the  process  of  identifying  the  most  accurate  ARIMA 
model  for  a  given  time  series.  These  functions  automate  the  process  of  fitting  and 
evaluating  multiple  ARIMA  models,  selecting  the  best  model  based  on  statistical  metrics 
such as AIC (Akaike Information Criterion). 
  The  auto_arima  function  in  the  pmdarima.arima  module  is  particularly  useful  for 
automatically identifying the optimal ARIMA model for a given time series. This function 
performs  a  stepwise  search  algorithm  over  multiple  ARIMA  models,  testing  each  model's 
performance  against  the  available  data.  The  function  then  returns  the  ARIMA  model  with 
the  lowest  AIC  value,  indicating  the  best  model  for  predicting  future  values  of  the  time 
series. 
 
Performing stepwise search to minimize AIC 
 
ARIMA(2,1,2)(1,1,1)[12]             : AIC=186.023, Time=0.31 sec 
ARIMA(0,1,0)(0,1,0)[12]             : AIC=206.697, Time=0.02 sec 
ARIMA(1,1,0)(1,1,0)[12]             : AIC=187.692, Time=0.05 sec 
ARIMA(0,1,1)(0,1,1)[12]             : AIC=inf, Time=0.11 sec 
ARIMA(2,1,2)(0,1,1)[12]               : AIC=184.636, Time=0.31 sec 
ARIMA(2,1,2)(0,1,0)[12]             : AIC=206.953, Time=0.06 sec 
ARIMA(2,1,2)(0,1,2)[12]             : AIC=186.049, Time=0.39 sec 
ARIMA(3,1,3)(0,1,1)[12]             : AIC=188.291, Time=0.35 sec 
ARIMA(2,1,2)(0,1,1)[12]             : AIC=inf, Time=0.31 sec 
 
Best model:  ARIMA(2,1,2)(0,1,1)[12] 	
â¢  2 the number of autoregressive (AR) terms 
â¢  1 differencing operation (to make the time series stationary) 
â¢  2 moving average terms 
â¢  1 seasonal autoregressive term 
â¢  1 seasonal differencing operation 
â¢  1 seasonal seasonal moving average term 
â¢  A seasonal cycle of 12 (i.e., monthly data) 	
 
  This model is used to forecast time series data that exhibit seasonal patterns. 
 
  Train  model  used  is  auto_arima(train,  trace=True,  error_action='ignore', 
suppress_warnings=True) 
 
   After  fitting  the  model  using  training  data,  it  is  essential  to  evaluate  the  model's 
performance  using  test  data.  The  graph  shown  the  predicted  values  of  test(orange)  and 
actual values (blue).

181 	
Compendium of Project Reports 
 
 	
 	
 	
Graph of actual predicted test data 	
 
Performance testing  
 
  RMSE  stands  for  Root  Mean  Squared  Error,  which  is  a  commonly  used  metric  to 
evaluate  the  performance  of  a  time  series  forecasting  model.  It  measures  the  difference 
between the predicted and actual values of the time series. 
  This  error  value  is  calculated  by  taking  the  square  root  of  the  average  of  the 
squared differences between the predicted and actual values.  
  A  lower  RMSE  value  indicates  that  the  model  is  better  at  predicting  the  future 
values of the time series. Here RMSE is 3.27. 
 	
Predicted CPI using SARIMA Model	 	
 
 
  
 
 
 
 
 
 
 
 	
Graph of future data 	
                          
  LSTM Model 
  
Data pre-processing 
 
  Data  pre-processing  is  a  crucial  step  in  data  analysis  and  machine  learning.  It 
involves  cleaning  and  transforming  raw  data  into  a  format  that  is  suitable for  analysis  and 
modeling.  One  common  pre-processing  step  is  to  fix  the  date  column  as  the  index 
column.In  time  series  analysis,  the  time  dimension  is  critical  and  often  serves  as  the  basis 
for forecasting and modeling. By setting the date column as the index column, we make it 
easier to slice, filter, and manipulate the data by date. 
 
 
 
Train Test  Split  	
Date              Predicted future values 	
2022-10-01                  195.011124 
2022-11-01                 195.951605 
2022-12-01                 197.103139

182                  	Compendium  of  Project  Reports	 	
 
 
  Splitting  a  dataframe    into  training  and  test  sets,  The  "train_test_split"  function 
from  the        sklearn  library  is  being  used  to  split  the data  into  two  sets:  80%  of the data  is 
being  assigned  to  the training  set    and  20%  is  being  assigned  to  the  test  set.  The "shuffle" 
argument is set to "False. 
 	
 	
 
 	
Train Test Split of data	 	  
Prepare Data 
 
  Prepare the data for a time series prediction problem with a look-back window size 
of 3. The data is converted into input (X) and output (y) pairs. X represents the window of 
previous n_features (3) time steps, while y represents the next time step following X.  
Reshapes the training and test data into a 3-dimensional format, which is required as input 
for LSTMs models. 
 
  In time series prediction problems, the input data is often transformed into input (X) 
and output (y) pairs. Define a look-back window size of 3, where X represents the window 
of  the  previous  3  time steps,  and  y  represents  the  next  time  step  following  X.  This  allows 
us to use the previous values to predict future values. 
 
  When  using  LSTM  models  for  time  series  prediction,  we  need  to  reshape  the  data 
into  a  3-dimensional  format  that  is  suitable  for  the  LSTM  input  shape.  The  3  dimensions 
are  (samples,  time  steps,  features),  where  samples  refer  to  the  number  of  observations  in 
the  dataset,  time  steps  refer  to  the  number  of  time  steps  in  each  input  sequence,  and 
features refer to the number of features in each time step. 
 
Train Model   	
â¢  model = Sequential() 
â¢  model.add(LSTM(8, input_shape=(1, look_back=3),)) 
â¢  model.add(Dense(1)) 
â¢  model.compile(loss='mean_absolute_error', optimizer='adamâ) 
â¢  model.fit(trainX, trainY, epochs=100, batch_size=12, verbose=1) 	
 
 
Test Model

183 	
Compendium of Project Reports 
 
  
testPredict = model.predict(testX) 
 
  LSTM  model  was  trained  using  the  Keras  deep  learning  library  in  Python.  The 
model  was  trained  with  a  look-back  window  size of  3,  where  X  represents  the  window  of 
previous  3  time  steps  and  y  represents  the  next  time  step  following  X.  The  training  data 
was fed to the model with 100 epochs and a batch size of 12. The loss function used for the 
model was mean absolute error and the optimizer was  Adam. The trained  model was then 
tested on the test data. 
 
Validation of Model 
 
  In  order  to  optimize  the  performance  of  the  LSTM  model,  we  experimented  with 
different  values  of  the  epoch  and  look  back  parameters.  After  testing  the  model  with 
different  values  of  these  parameters,  we  found  that  the  lowest  root  mean  squared  error 
(RMSE) was obtained with a look back window of 3 and 100 epochs.  
 
  LSTM model was trained and evaluated using the root mean squared error (RMSE) 
metric. The trained model was evaluated on both the training and test datasets. The RMSE 
of the model on the training set was found to be 1.55, indicating that the model performed 
reasonably well on the training data. Similarly, the RMSE of the model on the test set was 
found to be 1.27, indicating that the model was able to generalize well on the unseen data. 
The  lower RMSE  on  the  test  set  compared  to  the  training  set  suggests  that  the  model  was 
not overfitting the training data. Overall, the results suggest that the LSTM model was able 
to  effectively  capture  the  underlying  patterns  in  the  time  series  data  and  make  accurate 
predictions. 
 	
 	
 	
Graph of actual and predicted data 	
 
  
  	
   ------              	Train 	
  	 --------           	Predicted train 	
   --------------  Predicted test

184                  	Compendium  of  Project  Reports	 	
 
 	Both  ARIMA  and  LSTM  models  are  applied  to  the  dataset,  which  represents  the 	
Consumer  Price  Index  (CPI)  for  Kerala.  The  ARIMA  model  involves  a  series  of  steps, 
including data pre-processing, training/test data split, identifying and fitting the model, and 
evaluating  its  performance  using  metrics  such  as  RMSE.  The  best  ARIMA  model  is 
selected based on its AIC value and is then used to forecast future CPI values. The LSTM 
model  involves  a  similar  set  of  steps,  including  data  pre-processing,  train/test  data  split, 
and  model  fitting.  The  LSTM  model  is  a  neural  network-based  model  that  can  capture 
complex  temporal  dependencies  in  time  series  data.  The  LSTM  model  is  trained  and 
evaluated based on its ability to accurately predict future CPI values.  
The  study  comparing  ARIMA  and  LSTM  models  for  forecasting  consumer  price  index,  it 
was  found  that  the  LSTM  model  had  a  lower  RMSE  value  of  1.77  compared  to  the 
SARIMA  model's  RMSE  value  of  3.27.  This  indicates  that  the  LSTM  model  is  more 
accurate in predicting the future values of the time series data. 
   
CONCLUSION 
 
  Compared  to  traditional  time  series  models  such  as  ARIMA,  LSTM  models  can 
handle non-linear relationships and can automatically learn relevant features from the input 
data.  However,  they  require  significantly  more  data  and  computational  resources  for 
training and may suffer from overfitting if not properly regularized. Overall, LSTM models 
are  a  powerful  tool  for  time  series  analysis  and  are  widely  used  in  many  real-world 
applications. The study only  evaluated the performance of the ARIMA and  LSTM models 
on  one  dataset  .Using  a  longer  time  frame  could  potentially  provide  more  insights  into 
these patterns and trends.  
  In conclusion, the study demonstrated the effectiveness of both ARIMA and LSTM 
models in time series forecasting. However, the study is limited by the short duration of the 
dataset  used.  To  address  this  limitation,  future  studies  can  incorporate  longer  time  series 
with more historical data. Furthermore, other deep learning models and evaluation metrics 
can  be  explored  to  compare  the  performance  of  the  models.  By  using  more  robust 
evaluation techniques, the accuracy and reliability of the models can be enhanced. 
 
REFERENCES 
 	
[1]   Fundamentals of Applied Statistics , S.C.Gupta, V.K.Kapoor, Sultan Chand & Sons Publication. 
[2]    Dr  P  K  Sarangi,  Dr  Deepthi  Sinha,  Dr  Sachin  Sinha,  Dr  Meenakshi  Sharma    âForecasting  of 	
Consumer  Price  Index  Using  Neural  Networks  Modelâ.Innovative  Practises  in  Operations   
Management & Information Technology (ISBN 978-93-84562-11-3) 	
[3]   P.  K.  Sarangi,  D.  Sinha,  and  Mittal  (2021)  The  study  "Machine  Learning  Approach  for  the 	
Prediction  of  Consumer  Food  Price  Index"  .  2021  9th  International  Conference  on  Reliability, 
Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO) 	
  DOI: 10.1109/ICRITO51393.2021 
[4]   S  Zahara1  ,  Sugianto1and  M  B  Ilmiddaviq1  2020  Consumer  price  index  prediction  using  Long 	
Short  Term  Memory  (LSTM)  based  cloud  computing  Journal  of  Physics:  Conference  Series  1456 
(2020) 012022 doi:10.1088/1742-6596/1456/1/012022 	
[5]   Cheng  Yang  and  Shuhua  Guo  2021  âInflation  Prediction  Method  Based  on  Deep  Learningâ 	
Hindawi Computational Intelligence and Neuroscience Volume 20

185 	
Compendium of Project Reports
Population & Demographics

Agriculture

Industries

Socio-economic

Other data & insights

Price trends

RESEARCH IN STATISTICS

Population & Demographics

Agriculture

Industries

Socio-economic

Other Schemes/ Divisions

RECENT SURVEYS

Department Info

Contact

Directory

History

Organization

Announcements & Orders

Related Organisations

Acts and Rules

Public Portals

Official Login 1

Official Login 2

External Links

Latest Events

Miscellaneous

Employees' Portals

Site Administration

Publications

Category: General Publications

Transcript
