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.
Implications of the Growth of Population in Kerala 1964
Published Date : 18, Mar 1965
Implications of the Growth of Population in Kerala 1964
Ecostat Bulletin April 2025
Published Date : 22, May 2025
Ecostat Bulletin April 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of March 2025
Ecostat Bulletin March 2025
Published Date : 21, May 2025
Ecostat Bulletin March 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of February 2025
Ecostat Bulletin February 2025
Published Date : 20, May 2025
Ecostat Bulletin February 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of January 2025
Ecostat Bulletin January 2025
Published Date : 19, May 2025
Ecostat Bulletin January 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of December 2024
Ecostat Bulletin December 2024
Published Date : 17, May 2025
Ecostat Bulletin April 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of November 2024
Ecostat Bulletin November 2024
Published Date : 16, May 2025
Ecostat Bulletin April 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of October 2024
Ecostat Bulletin September 2024
Published Date : 14, May 2025
Ecostat Bulletin September 2024- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of August 2024
Ecostat Bulletin August 2024
Published Date : 09, May 2025
Ecostat Bulletin August 2025- Programmes and events conducted by Department of Economics & Statistics, SASA and Kerala State Statistical Commission during the month of July 2024
MORE PUBLICATIONS
Transcript

G
o
v
e
r
n
e
m
n t   o f   K e r a l a
1
9
5 1
R
E
P
O
R
T
 
O
N
 
P R E L I M I N A R Y
C
R
O
P
 
C
U
T
T
I
N
G
 
S
U R V E Y   O N   P A D D Y
I
N
 
P
U
N
J
A
 
F
I
E
L
D
S   O F   K U T T A N A D
A
N
D
 
N
E
I
G
H
B
O
U R I N G   A R E A S        
G
o
v
e
r
n
e
m
n
t
 
o
f
 
T
r a v a n c o r e -   C o c h i n
B
O
A
R
D
 
O
F
 
S T A T I S T I C S
C
e
n
t
r
a
l
 
R
e
s
e
a r c h   I n s t i t u t e
T
r
a
v
a
n
c
o
r
e   U n i v e r s i t y

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  11	th  Oct  2023  to  21	st  Dec  2023	
In collaboration with	
Department  of  Electronics  &  Information Technology  Kerala	
&	
Kerala  State  Information  Technology  Mission	
February 2024	
DIRECTORATE  OF  ECONOMICS  &  STATISTICS	
VIKAS BHAVAN, THIRUVANANTHAPURAM

Dr. RATHAN  U. KELKAR  	IAS                         	Electronics & Information Technology&	
   	 SECRETARY TO GOVERNMENT	           	        Environment Department
     	       Government Secretariat	
         Thiruvananthapuram, Govt. of Kerala	
   E-mail : secy.itd@kerala.gov.in	
   secy.envt@kerala.gov.in	
 Government of Kerala	
 0471-2336602	
          : 0471-2518444	
Date :17-02-2024	
Government funded programme of Capacity Building in Artificial Intelligence and Data Analytics is	
aimed at creating a set of employees in government departments to manage data for informed decision
making with the support of advanced IT and data science technology and also to integrate Machine
Learning and AI in government IT projects. It is learned that a group of 11 statistical staff  from Department
of Economics & Statistics has participated this programme conducted by Digital University Kerala in
2023. Last year also, 19 employees from the department have successfully undergone this course. This
course will be extremely helpful to statistical staff for expanding their IT knowledge and data processing
capability and thus to bring more robust decision supportive analytics, in their organisations.	
At this juncture, I would like to congratulate DUK Vice Chancellor Dr. Saji Gopinath and his team	
for organising this programme in a wonderful way. And, I would like to wish all the trainees in the Department
for expanding their official career and serving the Government more better way.	
I hope this compendium of reports will be a valuable reference for researchers and those who	
engaged in the field of Data Science and Official Statistics.	
MESSAGE	
Phone - Office

Kerala University of Digital Sciences	
    Innovation & Technology	
Technocity Campus, Mangalapuram	
Thiruvananthapuram- 695317	
                                        Date: 09-02-2024	
MESSAGE	
I am extremely happy to note that the Capacity Development Program in Artificial Intelligence and	
Data Analytics designed and delivered by Digital University for officials of Department of Economics and
Statistics (DES) has successfully completed its third edition. This unique program is designed to realise the
Keralaâs commitment to become a truly digital State where advanced technologies are used by Government
for improving citizen services, by imparting high end technology skills to Government employees in a
structured manner. It is heartening to note that the program supported financially by Department of Electronics
and IT, Government of Kerala has immensely benefitted officers of DES with proven expertise in Statistics
to hone their skills further, evidence of which are already visible in the excellent projects being initiated by
them as part of the program.  I am sure that DUK capacity building training has enhanced their capabilities
to build their own services and systems using cutting edge tools like AI & data analytics to address specific
challenges faced by the Department in a comprehensive manner.
Prof (Dr.) Saji Gopinath
       Vice Chancellor
Prof. Saji Gopinath

sherly@duk.ac.in	
We are indeed happy to present the third batch training #AIFORALL compendium, compiled by	
Economics and Statistics Department (DES) of Government of Kerala. The programme successfully com-
pleted with 45 days intensive training aiming of Creating Future talents in accordance with the changing
needs of Government. Upskilling Government employees for current and future technologies by leveraging
the potential of Artificial Intelligence to position themselves as leaders is the objective of the training pro-
gramme. The capacity building imparted to DES was a great success and the officials made a remarkable
contribution by implementing projects in critical areas of health, agriculture using their own department
data. The projects include Utilization and control of Electricity Consumption, Suicide its cause and reduc-
tion, Time series analysis of Price of Tomato, farm prices, Paddy yield, Milk Production etc using different
machine learning and deep learning models.	
Congratulations to all the participants who have successfully completed the course and projects and	
we are grateful to the Director Mr. Sreekumar B, Deputy Director Mr. D S Shibukumar & Deputy
Director Mr. Abhilash K for constant interaction and cooperation during 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 Col-
lege and Dr. Satheesh Kumar, Professor, Future Studies of University of Kerala.	
I am grateful to Dr. Saji Gopinath, Vice Chancellor and Dr. Mujeeb A, Registrar of Digital Univer-	
sity of Kerala for their 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 Programme.	
ACKNOWLEDGEMENT	
 Elizabeth Sherly
 Prof. Elizabeth Sherly   	                                                      Kerala University of Digital Sciences	
Distinguished Professor         Innovation & Technology	
Technocity Campus, Mangalapuram	
Thiruvananthapuram- 695317	
                                                                         Date :09-02-2024

Department of Economics & Statistics	
Vikas Bhavan P.O, Thiruvananthapuram- 695033
Phone - Office: 0471- 2305318, Fax: 0471- 2305317	
Phone - Res: 0471- 230090
Email: ecostatdir@gmail.com	
Website: www.ecostat.kerala.gov.in	
                                                                Date: 17-02-2024	
SREEKUMAR B	
           DIRECTOR	
 	
Government of Kerala	
PREFACE	
Information and Communication Technology sector is growing very fast and is very crucial for	
managing data. As the Department of Economics & Statistics is a data producing organisation, the application
of ICT and data science have much importance in timely dissemination of the data products with adequate
inference and interpretation.	
This compendium is a consolidation of project reports prepared and presented by the staff of the	
Department of Economics & Statistics who have undergone the Capacity building Programme in Artificial
Intelligence and Data Analytics which was conducted by Digital University of Kerala for a period of  45
days from 11	th October to 21	st December 2023. This is the 2	nd batch of AI training which was exclusively	
conducted for statistical pesonnel. In this batch, time a team of 11 officers of different cadre has succesfully
completed the AI course.	
This data science related short term course was very helpful to the Department of Economics &	
Statistics for transforming its conventional practice of analysis to the advanced analysing practices by
blending ICT capabilities with the domain knowledge of existing statistical personnel. I would like to
express my  gratitude  to Dr. Saji Gopinath, Vice Chancellor DUK, Dr. Elizabeth Sherly, Distinguished
Professor and Course Co-ordinator DUK, Sri. Surag M, Scientific Associate & Assistant Coordinator of
the AI batch, DUK for their sincere co-operation to consider staff, exclusively for DES in the batch and
this according to their convenience.	
I congratulate all participants who have successfully completed the course and submitted their project	
reports. I also appreciate the participants for their extra effort to attend the course in addition to their
normal duties. I extend  my sincere thanks to Additional Director (General), Sri. Santhoshkumar P. D.,
Deputy Director Sri. D. S. Shibukumar, and Deputy Director, Sri. Abhilash K, for their earnest efforts in
co-ordinating the programme  and make it a big success.	
I  hope that this report would be a future reference to all statistical personnel  engaged in the data	
analytics excercises and for transforming the State Statistical System (SSS) with the use of advanced data
analytics tools.	
Sreekumar B

Trainees of the Programme
1 Smt. Yamuna A.R, Deputy Director, Vital Statistics Division, Directorate
2 Smt. Deepa S.A, Deputy Director, Directorate of Agriculture & Farmerâs Welfare
3 Sri. Abdul Gafoor,  Assistant Director (IIP), Directorate
4 Sri. Sajin Gopi,  Assistant Director (State Income), Directorate
5 Sri. Sreekumar G, Research Officer, NSS Division, Directorate
6 Sri. Saju K, Research Assistant, EARAS Division, Directorate
7 Sri. Shibu B.T, Research Assistant, PPC Division, Directorate
8 Sri. Brijesh C.J, Statistical Assistant Grade I, PPC Division, Directorate
9 Sri. Adarsh R.S, Statistical Assistant Grade I, Directorate of Animal Husbandry
10 Kum. Lekshmi S, Statistical Assistant Grade II, EARAS Division, Directorate
11 Kum. Reshmi S, Statistical Assistant Grade II, EARAS Division, Directorate
I

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

INTRODUCTION	
The Department of Economics and Statistics is the Nodal Agency in Kerala for collection, compilation,	
analysis, objective interpretation and dissemination of data on all socio-economic sectors of Kerala economy.
The State has a robust statistical system in India to organize and conduct census and sample surveys 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 department is publishing on an average 30 reports of various categories in every year and the	
digital copies of the reports are available in the official website 	www.ecostat.kerala.gov.in	. The department	
has started digital publication from 2004. The website now contains around 1100 publication of various
category and the digital versions of them are available for free downloads. These publications includes
those published from 1955-56 but digitized recently. In addition, the department library contains many
more reports of statistical organizations in other States, Central and other State department publications
and academic books and references. Conventional methods of statistical analysis are usually performed in
the department for preparation of reports.	
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. Data Science provides latest techniques by
blending the ICT technology to analyse the data more efficiently and to provide a best result for effective
decision making in a quick and easier manner.	
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
2023-24. This programme of AI & DA is the third batch of the DUK with the funding of State IT department
and Kerala State IT Mission. For the Department of Economics Statistics, this AI course was the second
batch and the course was exclusively conducted for statistical people. The programme was conducted in
45 working days, started from 11 October 2023 and ended on 21 December 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
III

1
1	
IV
charges were borne by DUK. The to and fro transport facility to the participants has been arranged by
DES with the vehicle support of State Academy on Statistical Administration (SASA), Kerala. The 45
days course was successfully completed on 21 December 2023 with project presentation by each participant.
After the completion of the programme, project presentations were arranged before the directorate staff
during the months of January and February 2024 for famialiaring the data analysis done by the trainees and
to give  motivation to the rest of the staff.
Evolution of the programme	
DUK has called for nominations form DES for joining the programme of third batch AI & DA	
conducted by DUK with the budgetary support of Govt of Kerala. DES has sought willingness from the
directorate staff. Willingness was also sought from some staff working in line departments. 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. From the officers
who have expressed willingness to participate, 14 officials were 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. During last year in 2022-23, a total of 19 statistical personnel
in different cadre have successfully completed the AI & DA course conducted by DUK.
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 and RStudio and data analysis in R.
Project Assignment and Evaluation	
Two weeks project assignment was given to all trainees during the programme. Trainees have	
selected their own subjects and datasets for analysis and project preparation. Majority of the trainees used
DES datasets for data analysis and project preparation.

1	
Datasets used for project preparation	
The course participants have selected their own datasets according to their area of interest. Majority	
of the trainees have considered department datasets for analysis. The details of datasets used for analysis
in this project report are mentioned below.	
Electricity consumption data of the household survey conducted in Cherukunnu Grama Panchayat	
ion Kannur district, milk production data of Integrated Sample Survey of Animal Husbandry Department,
Farm price of coconut collected and published by DES on monthly basis, data of Sample Check study
conducted for field level correction in methodology under EARAS survey, production statistics of tapioca
in Kerala under EARAS survey, data collected for preparation of Consumer Price Index for Rural/ Urban
and Combined, Cost of Cultivation survey data, data on suicides in Kerala available under State Crime
Records Bureau, Monthly Percapita Consumption Expenditure under National Sample Survey and Civil
Registration data of deaths in Kerala for the past years are the datasets 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), Seasonal ARIMA (SARIMA), Artificial Neural Networks (ANN), Support
Vector Machines (SVM), Random Forest etc. are the methods used by the trainees for analyzing and
forecasting data.
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. The department already has taken steps to
set up a Data Analytics Unit in the Directorate. The Data Analytics Unit in the directorate is presently
having a team of five officers in different cadre, all have obtained AI training during the last batch of AI &
DA conducted by Digital University Kerala. Definitely this kind of initiates will lead to improve the data
analytics practices and research activities . Data Analytics Wing is essential to think like a data scientist and
analyze 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 for
effective planning and informed decision making. DES is one of the prime departments as a data provider
for planning purpose. With the support of Data Science, the department can transform itself through the
advanced level of capacity building programmes.
V

1

Sl No  Project Title  Page No.	 	
1 	
Analyzing Features of Residential Electricity Consumption 
through AI Approaches 	
Sri. Abdul Gafoor	 	
1 	
2 	
Forecasting Milk Production Trends in Kerala Using AI 
Model 	
Sri. Adarsh RS	 	
19 	
3 	
AI Model for Forecasting Farm Price of Coconut in Kerala 	
Smt. Deepa S.A.	 	
29 	
4 	
Forecasting Paddy Yield Using Sample Check Data 	
Kum. Lekshmi S	 	
41 	
5 	
Analysis and Forecasting of Tapioca Production in Kerala 	
Kum. Reshmi S	 	
49 	
6 	
Consumer Price Index(R/U/C) - A Machine Learning 
Approach to Cluster The Markets 	
Sri. Sajin Gopi	 	
57 	
7 	
Prediction of Farm Wholesale Price of Paddy - Cost of 
Cultivation as Predictor 	
Sri. Saju K	 	
71 	
8 	
An Analysis of Suicide in Kerala for the Past 26 Years 
(1996-2021) 	
Sri. Shibu B.T. & Sri. Brijesh C.J.	 	
81 	
9 	
Analysis and Prediction of Monthly Per Capita Expenditure 
(MPCE) of a Family in Kerala 	
Sri. Sreekumar G	 	
101 	
10 	
Time Series Forecasting Study for Death (1962-2020) Using 
AI 	
Smt. Yamuna A.R.	 	
113 	
INDEX  OF  PROJECT  REPORTS	
VI

1

1 	
 	
 
 
 
 	
 
 
 
 	
Analyzing Features of Residential Electricity 	
 	
Consumption through AI Approaches	
 	
 	
 	
 
 	
Submitted by 	
Sri. Abdul Gafoor,  	
Assistant Director

2 	
 	
Executive Summary 	
 	
This  report  presents  a  comprehensive  analysis  of  residential  electricity 	
consumption using advanced AI approaches. The research uses advanced methods to 
figure out which factors have a significant impact on electricity usage. By leveraging 
artificial  intelligence,  we  aim  to  uncover  patterns,  correlations,  and  insights  that 
traditional methods may overlook. 	
Residential electricity consumption is a multifaceted phenomenon influenced by 	
numerous  factors,  and  this  report  delves  into  a  comprehensive  analysis  utilizing 
advanced  artificial  intelligence  (AI)  approaches.  The  aim  is  to  unravel  intricate 
patterns,  correlations,  and  insights  that  traditional  methods  often  fail  to  capture.  By 
leveraging  the  power  of  AI,  this  study  seeks  to  enhance  our  understanding  of  the 
dynamic interplay between various features and their impact on residential electricity 
consumption. 	
From  our  investigation,  we  derive  the  linear  equation: 	Y=74.033  X	1  +108.131   	
X	2  +  349.248  X	3 +  0.145    X	4 +  165.839  X	5   +  111.702  X	6   +  5.438  X	7  +  17.403  X	8 + 	
146.187  X	9 +133.309  X	10 	+  119.654  X	11  	-  141.982  X	12	+C     	Where 	  X1 = Number of 	
Fan	,  X	2  =  Family  size      , 	X3  =  AC(1/0)*    , 	X4  =  Building  Area    , 	X5  =    Religion  2 	
(Islam)(1/0)  *  ,   	X  6 	  =  Income  from  Pravasi(1/0)  *, 	 X7  =  Number  of  LED, 	X8  = 	
Number  of  CFL, 	X9  =    Number  of  Laptops, 	X10	  =  Number  of  Tablet, 	X11	  =  Income 	
from  Business(1/0)  *,   	X12	  =    Floor  type  (Cement,Tiles)(1/0)  *  ,C    is  a  constant  and 	
here  C=0,	      *	AC,  Religion2  (Islam),  Income  from  Pravasi,  Income  from  Business, 	
and  Floor  type  (Cement,  Tiles)  are  binary  variables.  This  means  they  take  the  value 
of 1 if the corresponding facility is available, and 0 if it is not available. 	
 This  equation  represents  the  relationship  between  these  factors  and  residential 	
electricity consumption. Each coefficient (e.g., 74.03, 108.13) signifies the impact of 
the  corresponding  variable  on  the  predicted  electricity  consumption.  The  binary 
variables  (1/0)  indicate  the  presence  (1)  or  absence  (0)  of  certain  features,  such  as 
AC, Religion 2 (Islam), Income from Pravasi, and Floor type (Cement, Tiles). 	
The  findings  from  this  analysis  serve  as  a  valuable  basis  for  informed  decision-	
making  within  the  energy  sector.  It's  crucial  to  recognize  that  the  dataset  originates 
from census data, specifically focusing on Cherukunnu Gramapanchayath in Kannur. 
However,  it  may  not  encompass  all  aspects  of  the  entire  population,  offering  a 
glimpse into the features relevant to this specific area. The R-squared value of 49.19% 
in  the  regression  model  indicates  that  approximately  49.19%  of  the  variability  in 
electricity  bills  (ElecBill)  can  be  explained  by  the  included  independent  variables. 
While this measure gauges the model's fitness, it's important to note that a portion of 
the variability remains unaccounted for.

3 	
 
1.	 	Introduction: 
Residential  electricity  consumption  represents  a  critical  facet  of  modern  energy 	
dynamics,  and  understanding  the  underlying  factors  influencing  consumption  patterns  is 
paramount for effective energy management. In this report, we embark on a comprehensive 
exploration  of  residential  electricity  consumption,  leveraging  advanced  Artificial 
Intelligence  (AI)  approaches  to  discern  nuanced  relationships  and  uncover  insights  that 
transcend traditional analytical methods. 	
The significance of this analysis lies in the recognition of the evolving landscape of 	
energy  consumption,  shaped  by  diverse  socio-economic  factors,  technological 
advancements,  and  changing  lifestyles.  As  we  transition  towards  more  sustainable  and 
efficient  energy  practices,  a  deep  understanding  of  the  features  impacting  electricity 
consumption becomes imperative. Unveiling the Synergy of Floor Area, Family Dynamics, 
and Beyond Provide a concise overview of how AI is applied to comprehend the impact of 
specific Features on residential electricity consumption   	
In  conducting  this  analysis,  a  robust  dataset  involving  a  socio-economic  survey 	
conducted  in  November  2021  at  Cherukunnu  Gramapanchayath  in  Kannur  District.  A 
dataset  encompassing  information  from  3800  households  was  meticulously  curate, 
covering  both  household  and  individual  details.  This  comprehensive  dataset  forms  the 
basis  for  the  AI-powered  investigation  outlined  in  this  report.  The  survey  involves 
gathering the electricity bills for each residential house over a two-month period. 	
To  ensure  the  reliability  and  accuracy  of  the  analysis,  rigorous  data  preprocessing 	
techniques  were  applied.  This  involved  addressing  missing  values,  encoding  categorical 
variables, and creating a clean and standardized dataset suitable for advanced AI modeling. 	
The heart of the analysis lies in the exploration of influential features on residential 	
electricity  consumption.  Leveraging  machine  learning  algorithms,  regression  models  were 
employed  to  quantify  relationships  and  predict  consumption  patterns.  Techniques  such  as 
Ordinary  Least  Squares  (OLS),  linear  regression  and  Ridge  regression  were  utilized  to 
provide a nuanced understanding of how various features contribute to electricity usage. 	
Recognizing  the  potential  impact  of  outliers  on  model  performance,  the  analysis 	
includes  outlier  removal  techniques,  specifically  employing  box  plot  analysis  and  Z 
threshold.  	
In  assessing  the  multicollinearity  among  features,  Variance  Inflation  Factor  (VIF) 	
analysis  was  conducted,  offering  insights  into  potential  redundancies  and  correlations  that 
might affect the stability of the regression models. 	
Principal  Component  Analysis  (PCA)  was  incorporated  to  reduce  dimensionality 	
and  identify  patterns  in  the dataset.  Cumulative  explained  variance ratios  were  scrutinized 
to  determine  the  optimal  number  of  principal  components  contributing  to  electricity 
consumption patterns. 	
The  results  and  interpretations  obtained  from  these  analyses  provide  a  nuanced 	
understanding  of  the  features  influencing  residential  electricity  consumption.  Notably,  the 
Linear Regression model emerged as a powerful tool, offering comprehensive insights into 
the relationships within the dataset. 	
In  closing,  this  report  stands  as  a  testament  to  the  integration  of  AI  approaches  in 	
unraveling  the  intricacies  of  residential  electricity  consumption,  providing  actionable 
insights for stakeholders in the energy sector and beyond.

4 	
 	
2.	 	Objective of the Analysis: 	
 
1.	 	Identify  Influential  Features:	 Uncover  and  quantify  the  features  within 	
households that significantly impact electricity consumption	. 	
2.	 	Leverage  AI  Techniques:	 Harness  the  power  of  AI,  including  machine 	
learning  algorithms,  to  analyze  complex  relationships  and  patterns  within  the 
dataset. 	
3.	 	Inform  Decision-Making:	 	Provide  actionable  insights  for  stakeholders, 	
policymakers,  and  energy  professionals  to  make  informed  decisions  for 
optimizing residential electricity consumption.	 	
4.	 	Contribute  to  Sustainable  Practices:	 	Facilitate  the  transition  towards 	
sustainable energy practices by understanding the dynamics of electricity usage 
in residential settings. 
 	
3.	 	Data Collection: 	
 
The  study  utilized  a  robust  dataset  that  include  a  socio-economic  survey 	
conducted  in  November  2021  at  Cherukunnu  Gramapanchayath  in  Kannur  District. 
Information  from  3800  households  was  gathered  through  census  methodology.	 The 	
survey  specifically  concentrated  on  gathering  details  related  to  households  and 
amenities.  Additionally,  individual  information  was  also  collected.  focusing  on  both 
household  and  individual  details.  This  dataset  provides  a  rich  foundation  for  AI-
driven  analysis.  This  dataset  includes  a  wealth  of  information,  ranging  from 
household  demographics  to  specific  details  about  amenities  and  appliances.  The 
comprehensive  nature  of  the  survey  ensures  a  holistic  understanding  of  the  factors 
contributing to electricity bill. 	
 	
a.	 	Dataset 	
The  provided  dataset  includes  37  parameters  and  consists  of  3785  entries  after 	
the removal of missing values.

5 	
 	
b.	 	All Parameters: 	
]'A	1_ID',  'Ward',  'Team',  'SrlN',  'ClosStat',  'HsNo',  'Catag',  'Relegn','famMemb', 	
'RationType',  'IncmSrce',  'HousType',  'ResidYeras','HouseAge',  'RoadType', 
'NewsPaper', 'Ownership', 'FloorTp', 'WallTp','RoofTp', 'SqrFeet', 'ElecBill', 'GasYN', 
'GasCyl',  'ElecYN',  'SolarYN','ACYN',  'NonStar',  'LEDN',  'CFLN',  'FilamentN', 
'FanN', 'ToiletN', 'Desktop', 'Laptop', 'TabN', 'TVYN'] 	
c.	 	Numerical Parameters:  
['Ward', 'SrlN', 'HsNo', 'famMemb', 'ResidYeras', 'HouseAge', 'NewsPaper', 'SqrFeet', 
'ElecBill',  'GasCyl',  'NonStar',  'LEDN',  'CFLN',  'FilamentN',  'FanN',  'ToiletN', 
'Desktop', 'Laptop', 'TabN',] 	
d.	 	Categorical Parameters:  
['Team',  'ClosStat',  'Catag',  'Relegn',  'RationType',  'IncmSrce',  'HousType', 
'RoadType', 'Ownership', 'FloorTp', 'WallTp', 'RoofTp', 'GasYN', 'ElecYN', 'SolarYN', 
'ACYN', 'TVYN'] 	
4.	 	Data Preprocessing: 	
Data  preprocessing  is  a  crucial  phase  in  our  analysis,  acting  as  the  foundation  for 	
accurate  and  meaningful  insights.  This  process  involves  a  series  of  steps  to  clean, 
transform, and organize the raw data, ensuring that it is suitable for advanced AI modeling. 
The overarching goal is to address potential issues, such as missing values, scale disparities, 
and  categorical  data  representations,  that  could  impact  the  quality  and  reliability  of  the 
subsequent  analyses.  Prior  to  further  analysis,  the  following  10  parameters  were 
intentionally removed to mitigate their influence on electricity consumption: 	
ï· 	['A1_ID',  'Ward',  'Team',  'SrlN',  'ClosStat',  'HsNo',  'RoadType',  'GasCyl',  'ElecYN', 
'ToiletN']	  	
a.	 	Handling Missing Values: 
The absence of data can create biases and diminish the efficiency of AI models. In this 	
dataset,  12  households  lacked  electrification,  and  these  records  were  excluded. 
Subsequently,  robust  imputation  methods  were  utilized  to  handle  missing  values  in  the 
dataset,  specifically  for  variables  like  Electricity  Bill  ('ElecBill')  and  Building  Area 
('SqrFeet').  Techniques  such  as  mean  imputation  were  applied,  taking  into  consideration 
the characteristics of the missing data. 
b.	 	Encoding Categorical Variables: 
Many  machine  learning  algorithms  require  numerical  input,  necessitating  the 	
transformation  of  categorical  variables.  Techniques  such  as  one-hot  encoding  or  label 
encoding  were  employed  to  convert  categorical  data  into  a  format  suitable  for  model 
training. This ensures that the algorithms can effectively interpret and utilize these features. 
The categorical parameters yet to undergo one-hot encoding are as follows: 	
ï· 	['Catag',  'Relegn',  'RationType',  'IncmSrce',  'HousType',  'Ownership',  'FloorTp', 
'WallTp', 'RoofTp', 'GasYN', 'SolarYN', 'ACYN', 'TVYN'] 	
Following  the  conversion  of  these  categorical  parameters  into  Boolean  parameters 	
using one-hot encoding, this process involves generating dummy columns. 
 (As shown in the provided code)

6 	
 	
Then the   data set includes 68 parameters and consists of 3785 entries. Dataset given 	
below: 	
 	
Subsequently, transforming them into binary parameters 	
 	
The resulting 68 parameters are as follows:  	
ï· 	['famMemb',  'ResidYeras',  'HouseAge',  'NewsPaper',  'SqrFeet',  'ElecBill',  'NonStar', 
'LEDN',  'CFLN',  'FilamentN',  'FanN',  'Desktop',  'Laptop','TabN',  'Catag_1', 
'Catag_3',  'Relegn_1',  'Relegn_2',  'Relegn_3','RationType_1',  'RationType_2', 
'RationType_3',  'RationType_4',  'RationType_5',  'IncmSrce_1',  'IncmSrce_2', 
'IncmSrce_3', 'IncmSrce_4','IncmSrce_5', 'IncmSrce_6', 'IncmSrce_7', 'IncmSrce_8', 
'IncmSrce_9', 
'IncmSrce_10','IncmSrce_11','IncmSrce_12','IncmSrce_13','IncmSrce_14','HousTyp
e_1','HousType_2','HousType_3','Ownership_1','Ownership_2','Ownership_3','Floo
rTp_2','FloorTp_3','FloorTp_4','FloorTp_5',  'FloorTp_6',  'FloorTp_7',  'WallTp_1', 
'WallTp_2',  'WallTp_3',  'WallTp_4',  'WallTp_5',  'RoofTp_2',  'RoofTp_3', 
'RoofTp_4',  'RoofTp_5',  'RoofTp_6',  'GasYN_1',  'GasYN_2',  'SolarYN_1', 
'SolarYN_2', 'ACYN_1', 'ACYN_2', 'TVYN_1', 'TVYN_2']  	
c.	 	Perfect co-linearity 	
Each  categorical  variable  is  expanded  into  multiple  variables  based  on  the  number 	
of codes or groups it contains, and collectively, these variables exhibit mutual co linearity. 
Subsequently,  each  categorical  variable's  last  parameter  was  removed  due  to  mutual  co 
linearity.

7 	
 	
ï· 	['Catag_1','Relegn_3','RationType_5','IncmSrce_14','HousType_3','Ownership_3','F
loorTp_7','WallTp_5','RoofTp_6','GasYN_2','SolarYN_2','ACYN_2','TVYN_2']  	
After  removing  these  parameters,  there  are  now  55  parameters,  and  we  examined  the 
summary of the data frame, as displayed below: 
 	
 	
 
d.	 	Outlier Removal: 
Outliers,  which  are  data  points  that  exhibit  substantial  deviation  from  the  variables 	
such  as  Electricity  Bill  ('ElecBill')  and  Building  Area  ('Sqr.Feet')  in  the  dataset,  have  the 
potential to distort the training of models and affect predictions.  	
i. 	Box plot 	
Box  plots  are  particularly  useful  for  comparing  the distribution  of  different  groups 	
or  identifying  potential  outliers.  They  offer  a  visual  summary  that  goes  beyond  simple 
measures  of  central  tendency,  providing  a  clearer  understanding  of  the  data's  variability 
and  skewness.  Additionally,  box  plots  are  effective  in  conveying  the  symmetry  or 
asymmetry  of  a  dataset,  making  them  a  valuable  tool  in  exploratory  data  analysis  and 
statistical reporting. 	
We generate a box plot to visually represent the  distribution of electricity bills and 	
square footage.

8 	
 	
 	
ii.	 	Z-threshold for outlier removal 
 	
The  Z-threshold  for  outlier  removal  is  a  statistical  criterion  used  to  identify  and 	
eliminate  outliers  from  a  dataset.  It  involves  calculating  the  Z-score  for  each  data  point, 
which represents how many standard deviations a particular value is from  the mean of the 
dataset. A Z-threshold is then set and any data point with a Z-score beyond this threshold is 
considered  an  outlier  and  removed  from  the  dataset.  Adjusting  the  Z-threshold  allows  for 
flexibility  in  the  sensitivity  of  outlier  detection,  with  a  higher  threshold  being  more 
permissive  and  a  lower  threshold  being  more  stringent.  Upon  conducting  a  Python-based 
analysis to determine the optimal Z-threshold value, we identified the optimal threshold as 
4.5. Employing this optimal Z-threshold value allows us to effectively remove outliers and 
subsequently evaluate the summary of the data frame. 	
 	
 
5.	 Relationships among parameters 	
 	
i. 	Heat map	  	
Itâs  a  valuable  exploratory  data  analysis  (EDA)  tool.  It  provides  an  intuitive  and 	
comprehensive way to understand relationships within the data, enabling to make informed 
decisions during feature selection, model building, and other analytical tasks.

9 	
 	
ii.	 	Correlation Matrix  	
A  correlation  matrix  is  a  powerful  tool  for  understanding  the  interdependencies 	
between  variables,  guiding  decision-making  in  statistical  analysis,  and  providing  valuable 
insights  for  various  analytical  tasks.  A  correlation  matrix  is  a  square  matrix  that  displays 
the  correlation  coefficients  between  multiple  variables.  Each  cell  in  the  matrix  represents 
the  correlation  between  two  variables,  ranging  from  -1  (perfect  negative  correlation)  to  1 
(perfect positive correlation).Display the Correlation Matrix of the Top 25 Most Correlated 
Features is using the code and output given below:  	
 	
 	  	
Heat Map of the Top 25 Most Correlated Features

10 	
 	
 	
 
Correlation with Electricity Bill 
i. Positive Correlations: 	
There  is  a  positive  correlation  between  "Electricity  Bill"  and  variables  such  as  "FanN" 
(Number  of  Fan)-  (0.5033),  "SqrFeet"  (Building  area)-  (0.4656),  "ACYN_1" 
(Availability  of  AC)  -  (0.4767),  "famMemb"  (Family  size)-  (0.4813)  and  âRelegn_2â 
(Muslim)-(0.4549).  This  suggests  that  as  these variables  increase,  there  is  a tendency  for 
the "Electricity Bill" to increase as well. 	
ii.	 Moderate Correlations: 	
Moderate  positive  correlations  are  observed  with  variables  like  âIncmSrce_12â  (Income 
from  Pravasi)  -(0.2942)  ,  âLED  Noâ  -    (0.2669)  ,âRationType_3â  (Blue  Card)- 
(0.2320)  ,âRoofTp_5  â(Concrete)-  (0.2150)  ,âCFL  Noâ    (0.2162),  and  âFloorTp_6â 
(Marble)-    (0.1957).  These  relationships  are  not  as  strong  as  some  of  the  other  positive 
correlations. 	
iii.	 Negative Correlation: 	
'Relegn_1'  (-0.3811),  'FloorTp_2'  (-0.3050),  and  'RationType_1'  (-0.2218)  exhibit  a 
negative  correlation  with  'ElecBill.'  This  suggests  that  as  the  values  of  these  variables 
decrease, the 'ElecBill' tends to increase. 
 Understanding  these  correlations  is  crucial  for  identifying  influential  factors  that 
contribute to changes in electricity consumption. These correlations do not imply causation, 
and further analysis or experiments may be required to establish causal relationships. 
6.	 	Multi-collinearity 	
Multi-collinearity  refers  to  a  situation  in  multiple  regression  analysis  where  two  or  more 
independent  variables  in  a  model  are  highly  correlated,  making  it  difficult  to  isolate  the 
individual effect of each  variable on the dependent variable. This can lead  to unstable and 
unreliable  estimates  of  the  coefficients  and  their  standard  errors.  One  common  way  to 
assess multi-collinearity is through the Variance Inflation Factor (VIF).

11 	
 	
i. Variance Inflation Factor (VIF): 
 	
Variance  Inflation  Factor  (VIF)  analysis  was  conducted  to  assess  multicollinearity  among 
features. This provided insights into potential redundancies and correlations that might affect 
the stability of the regression models.	 	
VIF  measures  how  much  the  variance  of  the  estimated  regression  coefficients  increases 
when  your  predictors  are  correlated.  A  high  VIF  indicates  that  the  associated  independent 
variable  is  highly  correlated  with  the  other  variables  in  the  model,  and  this  can  create 
problems in the regression analysis. 
The formula for VIF for a particular independent variable is:       	
í µí±í µí°¼í µí°¹í µí± =	
1	
(1 â Ri	6) 	
Where: 
VIFi 	is	  the  Variance  Inflation  Factor  for  the  i	th 	variable. 	í µí±i	6 is  the	 R2 value  obtained  by 	
regressing the	 ith predictor against all the other predictors. 	
A  VIF  of  1  indicates  no  multi-collinearity.  VIF  values  greater  than  5  or  10  are  often 
considered high and may warrant further investigation.  
 	
 	
 	
 	
VIF values help assess the level of multicollinearity in a multiple regression model.VIF 	
value above 10 is often  considered high, indicating potential multicollinearity issues. Also 
a  Variance  Inflation  Factor  (VIF)  negative  value  is  unusual  and  technically  not  possible.

12 	
 
VIF values are expected to be non-negative.	 To fix problems like negative VIF values, try 	
improving  the  regression  model  by  adding  a  constant  term.  After  we  have  added  the 
constant  term,  use  the  following  codes  to  recalculate  the  VIF  for  each  variable.  This  will 
give you a more dependable measure of multicollinearity. 
 	
 	
 	
 âRoofTp_5â	  a	nd	 âRoofTp_3â	 have  VIF  values  above  10,  indicating  potential  issues 	
with multicollinearity  and âFloorTp_2â, âFloorTp_3â	 and 	âFloorTp_6â	 âhave moderate VIF 	
values,  suggesting  a  moderate  level  of  multicollinearity.	 To  tackle  this,  we  can  combine 	
âRoofTp_5â  and	 âRoofTp_3â	 as  âRoofâ	 and	 âFloorTp_2â,  âFloorTp_3â	 and  âFloorTp_6â 	
as	 âFloorâ	 helping to address the multicollinearity concerns. Following that, we proceed to 	
recalculate the VIF, and the results are presented below.

13 	
 	
These results show that, despite the changes made, 'Relegn_1' and 'Relegn_2' still have 	
VIF  values  above  3.  Additionally,  'Relegn_1'  shows  a  negative  correlation  with  the 
electricity bill in the correlation matrix. To address multicollinearity, we decide to remove 
the  variable  'Relegn_1'  from  the  model.  Finally,  the  dataset  includes  22  parameters, 
including the 'const' term. We are now proceeding with fitting the model. 	
7.	 	 Data Splitting: 	
The  dataset  was  split  into  training  and  testing  sets  to  facilitate  model  training  and 	
evaluation. This separation ensures that the model is trained on one subset of the data and 
evaluated  on  an  independent  subset,  providing  a  more  accurate  representation  of  its 
performance on unseen data.  	
8.	 	Model Fitting  	
Model  fitting  refers  to  the  process  of  training  a  statistical  model  based  on  the  given 	
data.  In  the  context  of  linear  regression,  Ordinary  Least  Squares  (OLS)  is  a  common 
method used for model fitting.  	
i. 	Regression Modeling: 	
Regression  models  were  employed  to  quantify  and  understand  the  relationships 	
between  influential  features  and  electricity  consumption.  Techniques  such  as  Ordinary 
Least  Squares  (OLS),  Linear  regression  and  Ridge  regression  were  utilized  to  model  and 
predict consumption patterns.	 	
a.	 Ordinary Least Squares (OLS)  	
In  linear  regression,  the  OLS  method  is  employed  to  find  the  line  (or  hyper  plane  in 	
multiple  dimensions)  that  minimizes  the  sum  of  the  squared  vertical  distances  (residuals) 
between the observed and predicted values.	 OLS model fitting aims to find the best-fitting 	
linear relationship between variables by minimizing the sum of squared differences. It is a 
foundational  method  in  linear  regression  analysis,  providing  estimates  for  the  coefficients 
that define the regression equation 
The Output is

14 	
 	
 	
 	
Interpretation  
1.	 	R-squared: 	
 The  R-squared  value  is  0.483,  indicating  that  the  model  explains  48.3%  of  the 	
variability in the dependent variable 'ElecBill.' 	
2.	 	Coefficients: 	
 The coefficients represent the estimated impact of each predictor variable on 'ElecBill.' 	
For  example,  'FanN'  has  a  coefficient  of  74.2687,  implying  that  for  each  unit  increase  in 
'FanN,' 'ElecBill' is expected to increase by 74.2687 units, holding other variables constant. 	
3.	 	Confidence Intervals: 	
The [0.025, 0.975] columns provide the 95% confidence interval for each coefficient. It 	
gives a range within which we can be reasonably confident that the true coefficient lies	. 	
4.	 	P-values	  	
 The  P-values  associated  with  each  coefficient  indicate  whether  the  corresponding 	
predictor  variable  is  statistically  significant.  A  P-value  less  than  0.05  is  often  considered 
significant. 
9.	 	Feature Analysis: 	
The  application  of  AI  approaches,  including  machine  learning  algorithms,  facilitated 	
the  identification  of  features  influencing  residential  electricity  consumption.  Notable 
features,  such  as  the  number  of  fans,  square  feet  area  of  the  building,  and  the  presence  of 
air conditioning, demonstrated significant correlations with electricity bills. 	
 	
i. 	Feature Engineering: 	
Feature  engineering  involves  creating  new  features  or  transforming  existing  ones  to 	
enhance  the  model's  ability  to  capture  patterns.  This  step  was  undertaken  judiciously,

15 	
 
exploring  interactions  and  combinations  of  features  that  could  provide  additional  insights 
into electricity consumption patterns. 	
ii.	 Principal Component Analysis (PCA): 	
PCA  was  employed  to  reduce  dimensionality  and  identify  patterns  in  the  dataset. 	
Cumulative  explained  variance  ratios  were  analyzed  to  determine  the  optimal  number  of 
principal  components.  PCA  is  a  powerful  tool  for  reducing  the  complexity  of  high-
dimensional  data,  aiding  in  visualization,  and  uncovering  the  underlying  structure.PCA  is 
employed  to  reduce  the  number  of  features  (dimensions)  in  a  dataset  while  retaining  as 
much  of  the  original  variability  as  possible.	 PCA  is  used  for  various  purposes,  including 	
noise  reduction,  feature  extraction,  and  visualization.  We  use  Principal  Component 
Analysis and find cumulative Explained Variance Ratio  	
 	
 	
 	
Based  on  the  given  conditions  and  considering  both  the  p-value  and  the  cumulative 	
explained variance ratio, we decided to remove 9 parameters. The final set of parameters is 
then determined after this removal. 	
 	
 	
10.	 	Model Fitting and Comparison Using OLS, Linear Regression, and Ridge Regression 	
We proceed to fit the model once more, employing the OLS Regression Model, Linear 	
Regression Model, and Ridge Regression Model. This is done to compare and evaluate the 
results  obtained  from  each  of  these  regression  techniques.  Table  displaying  Features  and 
Coefficients for Ridge Regression, Linear Regression, and OLS

16 	
 	
 	
Interpretation  
The  analysis  revealed  specific  features  that  play  a  crucial  role  in  influencing 	
residential  electricity  consumption.  Linear  regression  model  provided  a  comprehensive 
understanding of the complex relationships between these features and energy usage.   	
 	
 	
i. 	Key Features Impacting Residential Electricity Consumption in the Dataset 	
1.	 	Number of Fan 	
2.	 	Family size 	
3.	 	AC (Yes or No) 	
4.	 	Building Area 	
5.	 	Religion 2 (Islam) (Yes or No) 	
6.	 	Income from Pravasi (Yes or No) 	
7.	 	Number of LED 	
8.	 	Number of CFL 	
9.	 	Number of Laptops 	
10.	 	Number of Tablet 	
11.	 	Income from Business (Yes or No)

17 	
 	
12.	 	Floor type (Cement,Tiles) (Yes or No) 	
Based  on  the  findings,  we  can  calculate  residential  electricity  consumption  using  the 	
formula  	
 	
Y=74.033 X	1 +108.131   X	2 + 349.248 X	3 + 0.145  X	4 + 165.839 X	5  + 111.702 X	6  + 5.438 	
X	7 + 17.403 X	8 + 146.187 X	9 +133.309 X	10 	+ 119.654 X	11  	- 141.982 X	12	+C 	
     Where 	  X	1 = Number of Fan	                               X	2 = Family size 	
                   	X	3 = AC(1/0)*                                          	X	4 = Building Area         	
          	X	5 =  Religion 2 (Islam)(1/0) *               	X	 6 	 = Income from Pravasi(1/0) *        	 	
                  	X	7 = Number of LED                              	X	8 = Number of CFL 	
                  	X	9 =  Number of Laptops                       	X	10	 = Number of Tablet    	
                 	X	11	 = Income from Business(1/0) *       	X	12	 =  Floor type (Cement,Tiles)(1/0)*  	 	
C is a constant and here C=0 	
 	
*AC, Religion2 (Islam), Income from Pravasi, Income from Business, and Floor type 	
(Cement,  Tiles)  are  binary  variables.  This  means  they  take  the  value  of  1  if  the 
corresponding facility is available, and 0 if it is not available. 
ii.	 	Interpretation 
The coefficients in the linear regression model indicate the impact of some respective 	
feature in the electricity bill. Here's the interpretation of the key features: 
   	Number  of  Fans  (FanN)	: With all other factors  held constant, each  additional fan 	
in  a  household  is  associated  with  an  estimated  increase  of  approximately  74.03  rupees  in 
the electricity bill. 	
Family  Members (famMemb):	 With all other factors held constant, An additional 	
family  member  is  associated  with  an  estimated  increase  of  about  108.13  rupees  in  the 
electricity bill. 	
AC Availability (ACYN_1):	 Holding all other factors constant, the presence of an air 	
conditioner  (AC)  is  associated  with  a  significant  impact,  contributing  to  an  estimated 
increase of around 349.25 rupees in the electricity bill. 	
Square  Feet  Area  (SqrFeet):	  Keeping  all  other  factors  constant,  each  additional 	
square foot of living space corresponds to a modest increase of approximately 0.15 rupees 
in the electricity bill. 	
These interpretations help understand the direction and magnitude of the impact each 	
variable has on the electricity consumption prediction. 	
11.	 	Conclusion: 	
In  conclusion,  the  application  of  AI  approaches  has  significantly  enhanced  our 	
understanding  of  the  features  influencing  residential  electricity  consumption.  The  insights 
gained from this analysis provide a foundation for informed decision-making in the energy 
sector.  This  dataset  is  derived  from  census  data  and  specifically  pertains  to  Cherukunnu 
Gramapanchayath in Kannur. It's important to note that the dataset may not capture all the 
characteristics  of  the  entire  population.  Instead,  it  focuses  on  the  information  relevant  to 
Cherukunnu  Gramapanchayath,  providing  a  snapshot  of  the  features  and  variables  within 
this specific area. 	
When  interpreting  and  generalizing  findings  from  this  dataset,  it's  crucial  to  consider 	
its  limited scope and  applicability to the broader  population. The model's  R-squared value

18 	
 
of 49.19% indicates that approximately 49.19% of the variability in the dependent variable 
(ElecBill)  is  explained  by  the  independent  variables  included  in  the  model.  While  R-
squared  is  a  measure  of  the  model's  goodness  of  fit,  it's  important  to  acknowledge  that  a 
portion of the variability remains unaccounted for. 	
Regrssion modelâs explains about 49.19% of the factors influencing electricity bills in 	
Cherukunnu Gramapanchayath. The remaining 50.81% of the variability is not captured by 
the  model  and  could  be  attributed  to  other  factors  or  inherent  complexities  that  are  not 
considered in the current set of independent variables. 	
This  limitation  highlights  the  need  for  further  investigation,  potential  inclusion  of 	
additional relevant variables, or acknowledgment that there may be inherent variability that 
cannot be explained solely by the chosen predictors. 	
12.	 	  Acknowledgments: 	
  I  would  like  to  sincerely  express  our  gratitude  to  our  Director  for  affording  me  the 	
opportunity to participate in the training. Our heartfelt appreciation goes to all the teaching 
and non-teaching staff of DUK  who played an active role in both the project and training, 
enabling  the  success  of  this  analysis.  Their  invaluable  efforts  have  been  pivotal  to  the 
accomplishments of this project. 	
13.	 	References: 	
1.	 Predicting House Prices with Machine Learning â authored by Eric Kim and shared 
on kaggle. 	
14.	 	Source: 	
1.	 Dataset - Cherukunnu `Gramapanchayath Socio Economic Survey 2021data used 	
15.	 	Libraries and Tools Used 	
1.	 Scikit-learn: Used for machine learning model implementation. 	
2.	 Pandas: Utilized for data manipulation and analysis. 	
3.	 Matplotlib and Seaborn: Employed for data visualization. 	
4.	 TensorFlow: Applied for deep learning tasks. 	
5.	 Jupyter  Notebooks:  Used  as  the  primary  environment  for  code  development  and 
experimentation.

19 	
 	
 
 
 
 	
 
 	
Forecasting Milk Production Trends  	
in Kerala using AI Model 	
 	
 	
 	
 	
Submitted by 	
Sri. Adarsh R.S,  	
 Statistical Assistant Grade I

20 	
 
1.Introduction 	
The animal husbandry sector holds significant importance in Kerala, contributing to 	
various  aspects  of  the  economy,  nutrition,  and  livelihood.  Dairy  farming,  in  particular, 
plays  a  crucial  role  in  rural  development  by  providing  employment  opportunities  and 
supporting  local  economies.  It  serves  as  a  means  of  diversification  for  agricultural 
practices  in  Kerala,  allowing  farmers  to  integrate  dairy  farming  with  other  activities, 
thereby  reducing  dependence on  a  single  crop.  Moreover,  milk  forms  the  foundation  for  a 
range  of  diversified  dairy  products  such  as  ghee,  butter,  cheese,  and  various  traditional 
dairy  items,  adding  value  to  agricultural  produce  and  catering  to  diverse  consumer 
preferences. 
2.Objective  	
The objective of this project was to analyses thirty years of milk production data for 	
Kerala  and  develop  a  predictive  model  for  future  milk  production.  The  data  included 
season â wise and category milk production for cattle buffalo and goat from 1990 to 2021. 
3.Data set used 
The dataset was obtained from Integrated Sample Survey results, conducted from March to 
February. The survey period was divided into three seasons: 	
ï· 	Summer Season: 1st March to 30th June 	
ï· 	Rainy Season: 1st July to 31st October 	
ï· 	Winter Season: 1st November to 28th or 29th February

21 	
 
4.Data Analysis

22 	
 
 	
The  dataset  provided  detailed  information  on  milk  production  in  Kerala  across 	
different  seasons  (summer,  Rainy,  and  Winter)  and  categories  (Cattle,  Buffalo,  Goat)  for 
each  year.  Initial  exploration  revealed  increasing  trends  in  overall  milk  production,  with 
variations  across  categories  and  seasons.  There  was  also  seasonality  in  the  data,  with 
production being higher in the rainy season. Based the coefficient of variation Buffalo milk 
production  shows  the  highest  variability  in  seasonality  changes.  Cattle  milk  production 
shows  relatively  low  variability  in  seasonality  changes.  Goat  milk  production  falls  in 
between,  with  a  moderate  level  of  variability.  These  interpretations  provide  insights  into 
how each category of animals responds to seasonal changes in terms of milk production. 
5. Methodology  	
The  study  utilized  machine  learning  models  and  Python  programming  for  data 	
analysis. The 	SARIMA model	, an extension of the ARIMA model, was employed for time 	
series forecasting, considering the presence of seasonal patterns in the data. Components of 
SARIMA included Seasonal, Autoregressive, Integrated, and Moving Average components, 
each defined by specific parameters. 
Seasonal (S): This indicates the presence of seasonality in the data. 
Autoregressive (AR): This component models the relationship between an observation and 
a number of lagged observations (autoregression). 
Integrated  (I):  This  component  accounts  for  the  differencing  needed  to  make  the  time 
series stationary, i.e., to remove trends from the data. 
Moving  Average  (MA):  This  component  models  the  relationship  between  an  observation 
and a residual error from a moving average model. 
The SARIMA model is specified by three main sets of parameters: 
p, d, q: These parameters are for the non-seasonal components of the time series data. They 
represent  the  order  of  the  autoregressive  (p),  differencing  (d),  and  moving  average  (q) 
terms, respectively. 
P,  D,  Q:  These  parameters  are  similar  to  p,  d,  and  q,  but  they  operate  on  the  seasonal 
component of the time series. 
s: This parameter represents the number of time steps in each season

23 	
 
Check Stationarity using Augmented Dickey-Fuller Test 	
The  Augmented  Dickey-Fuller  (ADF)  test  is  a  statistical  test  used  to  determine 	
whether  a  unit  root  is  present  in  a  time  series  dataset.  A  unit  root  indicates  that  a  time 
series data has a stochastic trend, making it non-stationary. The null hypothesis of the ADF 
test  is  that  the  time  series  data  has  a  unit  root,  meaning  it  is  non-stationary.  The  alternate 
hypothesis is that the time series data is stationary, meaning it does not have a unit root. In 
the  case  of  our  time  series  p-value  =  0.10262423774638108>0.05,  indicates  that  time 
series  is  non-stationary.  After  applying  log  transformation  and  differencing,  p-value  = 
0.000095<0.05, the data became stationary. 	
   	
 
 
 
Model Fitting: 
We  took  80%  data  as  train  data  set  and  20  %  as  test  data  set.  Using  the  Auto_arima 
function in the Pdarima library, the best-fitted SARIMA model was determined. In general, 
lower  AIC  (Akaike  Information  Criterion)  value  indicate  a  model  better  fit.  The  selected 
model was ARIMA(3,0,1)(0,0,0)[4].

24 	
 
 
 	
Milk Production Actual vs Prediction using SARIMA Model 	
 	
 
 
 
 
Model Evaluation: 
The predictive model's performance  was  evaluated using Mean Absolute  Percentage Error 
(MAPE) and Mean Absolute Percentage Error (MAPE): 	
MAPE  measures  the  average  absolute  percentage  difference  between  actual  and 	
forecasted values, expressed as a percentage. It is calculated using the formula: 	
1
í µí±	

Í	
|í µí°´í µí±í µí±¡í µí±¢í µí±í µí±	Üâ	í µí°¹í µí±í µí±í µí±í µí±í µí±í µí± í µí±¡í µí±í µí±	Ü|	
|í µí°´í µí±í µí±¡í µí±¢í µí±í µí±	Ü|	
á
Ü@5	
 í µí±	 100	 	
Where,  n  is  the  number  of  observations  in  the  data  set, 	í µí°´í µí±í µí±¡í µí±¢í µí±í µí±	Ü is  the  actual  value 	
at time 	i and 	í µí°¹í µí±í µí±í µí±í µí±í µí±í µí± í µí±¡í µí±í µí±	Ü  is the predicted value at time 	i 	
Root Mean Squared Error (RMSE). 	
RMSE measures the square  root  of the average of the squared differences  between 	
actual and forecasted values.  
   The  obtained  MAPE  value  of  2.83%  indicates  a  minimal  average  percentage 
difference  between  predicted  and  actual  values,  while  the  RMSE  value  of  36.38  signifies 
the typical magnitude of errors. These metrics affirm the model's effectiveness in capturing 
underlying data patterns and providing accurate predictions. 
 
 
 	
Year	
Total Milk 	
Production	
( in 000 	
Prediction	
(using 	
SARIMA)	
Summer 2017 822.04 814.602475
Rainy 2017 889.06 890.94
Winter 2017 864.88 799.83423
Summer 2018 833.56 817.251008
Rainy 2018 881.03 881.580105
Winter 2018 834.09 807.125008
Summer 2019 833.95 819.538415
Rainy 2019 891.92 873.714269
Winter 2019 818.48 812.940531
Summer 2020 807.21 821.537516
Rainy 2020 894.83 867.134186
Winter 2020 831.84 817.58027
Summer 2021 808.27 823.296254
Rainy 2021 900.32 861.64602
Winter 2021 823.87 821.2811

25 	
 
 
Adequacy Check and Visualization: 
Q-QPlot 	
 	
The  adequacy  of  the  model  is  checked  by  using  a  Q-Q  plot.  A  Q-Q  plot  is  a  valuable 
visualization  to  assess  the  normality  of  the  model  residuals.  The  plot  compares  the 
quantiles of the residuals against the quantiles of  a theoretical normal distribution. The Q-
Q  plot  falls  approximately  along  a  straight  line,  it  suggests  that  the  residuals  follow  a 
normal distribution. 	
 	
The  line  graph  appears  to  show  a  mostly  horizontal  line around  the zero  line,  with 	
some  variability.  There  are  some  data  points  above  and  below  the  zero  line,  but  none 
deviate  far  from  it.  Overall,  the  image  suggests  that  the  residuals  from  the  time  series 
model being used are relatively small and evenly distributed around the zero line. This is a 
good indication that the model is fitting the data well and making accurate predictions.

26 	
 	
 	
The  density  plot  below  illustrates  the  distribution  of  residuals  obtained  from  our 	
predictive  model.  A  density  plot  offers  a  visual  representation  of  the  probability  density 
function of the residuals, enabling insights into their distributional characteristics 
 
Future Prediction

27 	
 	
 	
 
6.Tools and Libraries Used 	
ï· 	Pandas, Numpy- libraries used for data analysis and manipulation. 	
ï· 	Mathplotlib- libraries used for creating high-quality visualizations, including charts, 	
ï· 	plots, and graphs. 	
ï· 	Seaborn- data visualization library that is built on top of Matplotlib 	
ï· 	Statsmodels-library for statistical modeling and analysis 	
ï· 	Sklearn- library for machine learning 	
ï· 	Scipy-library for scientific computing and technical computing. 	
ï· 	Pmdarima- library for automatic time series forecasting with ARIMA models 	
7.Conclusion 	
In conclusion, the SARIMA predictive model effectively forecasts milk production 	
in  Kerala  based  on  historical  data.  The  model's  accuracy,  as  indicated  by  low  MAPE  and 
RMSE  values,  suggests  its  reliability  for  future  predictions.  Additionally,  the  adequacy  of 
the  model  was  confirmed  through  a  Q-Q  plot  analysis,  which  indicated  that  residuals 
follow a normal distribution. 	
However,  it's  essential  to  acknowledge  certain  limitations  inherent  in  this  study. 	
Firstly,  the  predictive  model's  accuracy  heavily  relies  on  the  quality  and  completeness  of 
the dataset used for training. Incomplete or inaccurate data could lead to biased predictions 
and undermine the model's effectiveness. Secondly, while SARIMA is a robust method for 
time series forecasting, it may not capture all complex underlying factors influencing milk 
production,  such  as  changes  in  environmental  conditions,  disease  outbreaks,  or  policy 
interventions. 	
Furthermore,  the  model's  performance  may  vary  when  extrapolating  beyond  the 	
observed  data  range,  especially  in  the  face  of  unforeseen  events  or  systemic  shifts  in  the 
agricultural  sector.  Additionally,  the  SARIMA  model  assumes  stationarity  and  may  not 
adequately account for structural changes or long-term trends in milk production patterns. 	
Despite  these  limitations,  the  project  underscores  the  importance  of  leveraging 	
advanced  statistical  methods  for  forecasting  and  planning  within  the  animal  husbandry 	
Year	
Future 	
Prediction	
(in 000 tonnes)	
Summer 2022 824.848021	
Rainy 2022 857.078835	
Winter 2022 824.231194	
Summer 2023 826.217732	
Rainy 2023 853.284715	
Winter 2023 826.580567	
Summer 2024 827.425327	
Rainy 2024 850.137245	
Winter 2024 828.449142	
Future Prediction using SARIMA

28 	
 
sector.  Future  research  could  explore  integrating  additional  data  sources  and  employing 
more sophisticated modeling. 
 References used  	
ï· 	Integrated  Sample  Survey  Report  from  1990  to  2021  from  Department  of  Animal 
Husbandry 	
ï· 	Compendium  of  Project  Reports  #AIFORALL  Capacity  Building  in  Artificial 
Intelligence & Data Analytics from Department of Economics & Statistics 	
ï· 	https://www.kaggle.com/code/sunaysawant/air-passengers-time-series-arima	 	
.

29 	
 
 
 
 
 
 
 
 
 
 
 
 
 	
 
 
 
 
 	
AI Model for Forecasting Farm Prices 	
of Coconut in Kerala 	
 	
 	
 
 	
 
 	
Submitted by 	
Smt. Deepa S.A	,  	
Deputy Director

30 	
 
1.	 Introduction 	
India  is  the  largest  coconut-producing  country  in  the  world  and  ranked third  in  the 	
list of global coconut-producing nations. Although Kerala is known as âthe land of kera 
(coconut)â,  the  present  scenario  with  respect  to  production  of  coconut  is  not 
promising. 	 In  Kerala  Interior  places  with  fertile  soils  and  plain  regions  also  give  good 	
growth  to  coconut  trees.  During  2021-22,  Kerala  had  the  largest  area  under  coconut 
(36%)  followed  by  Karnataka  (28%)  and  Tamil  Nadu  (21%).  But  during  this  period, 
Kerala was in the third position with respect to production (25%), with Karnataka in the 
first position (31%) and Tamil Nadu in second position (26%). 	
The  agricultural  sector  plays  a  pivotal  role  in  the  economic  landscape  of  Kerala, 	
with  coconut  cultivation  standing  out  as  a  cornerstone  of  the  state's  agrarian  economy. 
As  a  major  contributor  to  both  rural  livelihoods  and  the  overall  economy,  the  coconut 
industry's  sustainable  growth  is  essential  for  ensuring  the  well-being  of  countless 
farmers and stakeholders. 	
The price behaviour of coconut and its products has profound influence on the rural 	
economy  of  Kerala.  Large  variation  in  the  prices  of  coconut  and  coconut  oil  within  a 
year  is  the  major  problem  faced  by  farmers,  consumers  as  well  as  planners.  Farmers 
depending  on  income  from  coconut  should  get  remunerative  price  throughout  the  year. 
Hence analysis of time series data of farm price of coconut is of prime importance. 	
2.	 Scope and Objectives 	
The scope of this project encompasses the development, validation, and application 	
of  ARIMA  (Autoregressive  Integrated  Moving  Average)  and  SARIMA  (Seasonal 
ARIMA) models  for  forecasting the farm prices  of coconuts in the state of Kerala. The 
project  will  focus  on  analysing  historical  price  data,  identifying  temporal  patterns,  and 
incorporating seasonality to enhance the accuracy of predictions. The temporal nature of 
agricultural  markets,  influenced  by  factors  such  as  climate  conditions,  demand 
fluctuations,  and  harvesting  seasons,  makes  time-series  analysis  an  apt  approach  for 
forecasting coconut prices. 	
The primary objectives of the project are: 	
ï· 	To  develop  accurate  forecasting  model  to  predict  the  temporal  patterns  in  the 	
farm price of coconut 	
ï· 	Generate  timely  and  accurate  forecasts  for  future  coconut  farm  prices,  aiding 	
stakeholders  in  making  informed  decisions  related  to  cultivation,  trading,  and 
policy formulation. 	
ï· 	Contribute  to  the  reduction  of  financial  risks  for  coconut  farmers  by  offering 	
insights  into  potential  price  fluctuations  and  enabling  proactive  risk  management 
strategies. 	
ï· 	Provide  a  valuable  tool  for  farmers,  traders,  and  policymakers  to  make 	
informed  decisions  related  to  planting,  harvesting,  pricing,  and  market 
interventions.

31 	
 
3.	 Literature Review 	
An  article  named  âPrice  Behaviour  in  Indiaâs  Coconut  Sectorâ  by  Dr.  Prafulla  K. 	
Das. In this article, an attempt has been made to predict the prices of copra with the use 
of coconut oil prices; and the prices of coconut with the uses of prices of either copra or 
coconut oil in the same market or in different markets. 	
An  article  in  the  âInternational  Journal  of  Chemical  Studiesâ  named  as  âPrice 	
behaviour of coconut in major markets of Kerala: A time series analysisâ by Preethi VP, 
M.Sc. (Ag.) Student, Jessy Thomas K, Professor and Anil Kuruvila, Associate Professor, 
Department  of  Agricultural  Economics,  College  of  Horticulture,  Thrissur,  Kerala 
Agricultural  University.  In  this  journal  they  analysed  the price behaviour  of coconut  in 
major markets of Kerala (Alappuzha & Kozhikode) for two periods and the variations in 
price to trend, cyclical, seasonal and irregular fluctuations were calculated. 	
A  study  named  as  âPrice  forecasting  models  for  coconut  and  coconut  oilâ,  has 	
conducted by  Indraji KN, Kerala Agriculture University, Thrissur.  In this study aims to 
evaluate  different  time  series  forecast  models  for  prices  of  coconut  oil,  copra  and 
coconut and to suggest the suitable forecast models for all. 	
4.	 Data used for modelling and validation with data source 	
The  dataset  used  in  this  project  is  the  monthly  average  farm  wholesale  price  of 	
coconut  (Rs  /  100  numbers)  in  Kerala  spanning  from  January  2000  to  December  2020. 
The  data  was  collected  from  various  years  of  âPrice  Statisticsâ,  a  publication  of 
Department  of  Economics  and  Statistics.  The  department  has  been  collecting  the  Farm 
price  of  major  agricultural  commodities  produced  in  the  State  since  1954.  The  farm 
wholesale  price  is  being  collected  fortnightly  from  77  Taluk  Centres  in  the  State. 
District  Average  and  State  Average  of  Agricultural  commodities  is  generated  on 
monthly basis.  	
 	
The dataset contains 252 rows and few rows are given below. 	
Sl No 	Month 	Farm price 	
0 	2000-01 	534.00 	
1 	2000-02 	474.00 	
2 	2000-03 	419.00 	
3 	2000-04 	389.00 	
4 	2000-05 	340.00 	
... 	... 	... 	
247 	2020-08 	1756.25 	
248 	2020-09 	1813.08

32 	
 	
249 	2020-10 	1869.46 	
250 	2020-11 	1925.62 	
251 	2020-12 	2033.73 	
    252 rows Ã 2 columns 	
Visualization of farm price over time 2000 to 2020 	
 	
5.	 Methods and Methodology used 	
The  time  series  values  of  the  variable  namely  farm  price  of  coconut  in  Kerala 	
recorded at monthly interval is used in this study. 	
The  forecasting  of  farm  prices  is  a  critical  aspect  of  agricultural  management, 	
providing  insights  for  both  producers  and  stakeholders  in  navigating  market  dynamics. 
In this time series analysis, specifically the Auto Regressive Integrated Moving Average 
(ARIMA)  and  Seasonal  ARIMA  (SARIMA)  models  are  used  to  predict  farm  prices  of 
coconut in Kerala. 	
The general processes used for ARIMA models is the following: 	
ï· 	Gathered  historical  farm  price  data  from  various  Price  Statistics  of  Dept.  of 
Economics & Statistics and compiled the dataset covering a substantial timeframe 
to capture temporal patterns and trends 	
ï· 	Smoothed  outliers,  and  applied  time  series-specific  techniques  (differencing, 
logarithmic transformations) for stationarity. 	
ï· Utilized  EDA  techniques,  including  time  series  plots,  ACF,  and  PACF,  to  identify 
temporal patterns and seasonality.

33 	
 	
ï· 	Choose  ARIMA  and  SARIMA  models  for  their  ability  to  capture  temporal 
dependencies and seasonality. 	
ï· 	Fine-tuned  model  parameters  through  differencing  and  identification  of  AR  and 
MA components	. 	
ï· 	Split dataset into training and validation sets. 	
ï· 	Utilized metrics such as MAE for model accuracy assessment. 	
ï· 	Ensured models' robustness through rigorous validation processes. 	
ï· 	ARIMA and SARIMA models demonstrated effective forecasting capabilities. 	
ï· 	Validation metrics indicated reliable predictive accuracy. 	
ï· 	Use the fitted model to make future prediction of farm price. 
The seasonal decompose function is used for time series decomposition, where a 	
time  series  is  decomposed  into  its  trend,  seasonal,  and  residual  components  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. The 
seasonal decomposition chart is given below. 	 	
On  conducting  component  analysis  of  our  time  series,  we  can  understand  that 	
our time series has an increasing trend, seasonality and noise. So, SARIMA model time 
series forecasting is suitable. 	 	
SARIMA (Seasonal ARIMA) Model	 	
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

34 	
 	
moving  average  (MA)  model.  By  addressing  seasonal  pattern,  SARIMA  extends 
ARIMA to handle seasonality. 	
The  key  steps  involved  in  developing ARIMA  &  SARIMA  models  are given 	
below: 	
1. Selection of differencing order to achieve stationarity 	
Stationarity  is  a  desirable  property  for  time  series  data  because  it  simplifies  the 	
modelling process. A Stationary series is one whose statistical properties such as mean, 
variance, covariance, and standard deviation do not vary with time.	 Time series-specific 	
preprocessing  techniques,  such  as  differencing  and  logarithmic  transformations,  were 
applied  to  stabilize  variance  and  achieve  stationarity.  The  order  of  differencing  (d  in 
ARIMA  and  D  in  SARIMA)  represents  the  number  of  times  differencing  is  needed  to 
achieve stationarity. 	
Augmented  Dickey-Fuller  (ADF)  test  is  a  statistical  test  used  to  determine 	
whether a unit  root is present in a univariate time series dataset. The presence of  a unit 
root  indicates  that  the  time series  data is  non-stationary.  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. 	
After  first  differencing,  the  ADF  test  was  conducted  and  the  following  result 	
obtained: 	
  ADF Test Statistic = 	- 6.503673674983843 	
           p-value = 1.1452Ã10	â8	  	
Which is less than 0.05 and hence the time series is stationary. 
Here  we  had  done  one  time  differencing  to  make  the  data  stationary.  So,  the 	
value of parameter âDâ of the SARIMA model can be taken as 1. 
2. Identification of AR and MA components to capture temporal dependencies 
  Autoregressive (AR) and Moving Average (MA) components capture dependencies 
between  current  and  past  observations.  ACF  (Auto  Correlation  Function)  and  PACF 
(Partial  Auto  Correlation  Function)  analyses  are  used  to  identify  significant  lags  for 
inclusion in the model. ACF helps to identify AR component and PACF helps to identify 
MA component. SARIMA extends ARIMA to handle seasonality.  It  includes additional 
seasonal  parameters  (P,  D,  Q)  to  model  seasonal  dependencies.  Seasonal  orders  (P,  D, 
Q)  represent  the  order  of  the  seasonal  AR,  seasonal  differencing,  and  seasonal  MA 
components, respectively	. 	
Exploratory Data Analysis (EDA) was conducted to gain insights into the temporal 	
patterns  and  seasonality  inherent  in  the coconut  price data. Visualizations,  such  as  time 
series  plots,  autocorrelation  functions  (ACF),  and  partial  autocorrelation  functions 
(PACF),  were  employed  to  identify  potential  patterns  and  inform  the  choice  of  model 
parameters âPâ & âQâ.

35 	
 	
ACF and PACF plots for model order selection	 	
 	 	
The lag where PACF drops off can suggest the order of the AR component âPâ and the 	
lag  where  ACF  drops  off  can  suggest  the  order  of  the  MA  component  âQâ.  Use 
decomposition  techniques  to  identify  the  periodicity  of  seasonality  in  the  data.  In  this 
study, the season is a âmonthâ. 	 	
In  this  study,  we  get  the  best  SARIMA  model  with  seasonal  order  (P,  D,  Q,  seasonal 	
period) = (2, 1, 10, 12) and the fitted model is given below.	 	
Plot for actual and predicted farm price	 	
 	
Once  a  Seasonal  Auto  Regressive  Integrated  Moving  Average  (SARIMA)  model  is 	
fitted,  it's  crucial  to  assess  its  performance  and  validate  its  predictive  capabilities.  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:

36 	
 	
1
í µí±	

Í	
|í µí°´í µí±í µí±¡í µí±¢í µí±í µí±	Üâ í µí°¹í µí±í µí±í µí±í µí±í µí±í µí± í µí±¡í µí±í µí±	Ü|	
|í µí°´í µí±í µí±¡í µí±¢í µí±í µí±	Ü|	
á
Ü@5	
 í µí± 100	 	
Where,  n  is  the  number  of  observations  in  the  data  set, 	í µí°´í µí±í µí±¡í µí±¢í µí±í µí±	Ü is  the  actual  value  at 	
time 	i  and 	í µí°¹í µí±í µí±í µí±í µí±í µí±í µí± í µí±¡í µí±í µí±	Ü  is  the  predicted  value  at  time 	i.    In  our  case  MAPE  is  1.98,  that 	
indicates good accuracy in your forecasting model	.  	
A Root Mean Squared Error (RMSE) is another measure of the accuracy of SARIMA 	
model.  The  RMSE  represents  the  square  root  of  the  average  of  the  squared  differences 
between the actual and predicted values. Here RMSE= 53.99, which means the differences 
between  the  predicted  and  actual  values  are  around  53.99  in  the  same  unit  as  the  actual 
data. 	
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. A quantile-quantile (Q-Q) plot 
is used to assess the normality of residuals. It's a valuable diagnostic tool for understanding 
the distributional characteristics of residuals and assessing the model's assumptions. 	
 	
Here  the  points  are  approximately  along  with  the  diagonal  line  and  hence,  the 	
residuals are approximately normally distributed. 	
Autocorrelation  and  partial  autocorrelation  of  residuals  are  important  tools  in  the 	
analysis  of  time  series  data,  particularly  in  the  context  of  modelling  and  forecasting  using 
techniques  like  autoregressive  integrated  moving  average  (ARIMA)  or  seasonal  ARIMA 
(SARIMA)  models.	 To  check  for  autocorrelation  in  residuals,  the autocorrelation  function 	
(ACF)  and  partial  autocorrelation  function  (PACF)  plots  are  used.  Autocorrelation 
measures  the  correlation  between  the  residuals  at  different  time  points  and  partial 
autocorrelation  of  residuals  helps  identify  the  direct  relationship  between  the  residuals  at 
two different time points, excluding the influence of other lags.

37 	
 	
ACF and PACF plots of residuals 	
 	
 	
 	
From the ACF plot it is clear that correlation coefficients between residuals at any lag 	
except  0  and  12  are  inside  the  blue  band  (indicating  a  confidence  interval)  and  hence 
statistically  insignificant,  means  there  is  no  significant  auto  correlation  between  residuals. 
So,  we  can  conclude  that,  this  time  series  model  adequately  captures  the  temporal 
dependencies  in  the  data.  Hence,  we  can  make  predictions  using  this  model.  Using  the 
model, we got the forecasted the monthly farm price of coconut for next one year as;

38 	
 	
SARIMA future data forecast 	
 	
 
 	
The future data predicted for 12 months of 2021 is given below. 
 	
Month  Predicted Farm Price (in Rs) 
Jan -2021  1882.48	 	
Feb â 2021  1902.81	 	
Mar â 2021  1922.39	 	
Apr â 2021  1941.56	 	
May â 2021  1945.46	 	
June â 2021  1940.85	 	
July â 2021  1950.12	 	
Aug â 2021  1966.37	 	
Sep â 2021  1985.70	 	
Oct â 2021  2002.67	 	
Nov â 2021  2015.54	 	
Dec - 2021  2030.75

39 	
 
6. Packages and libraries used 	
The libraries and tools used in this project are given below. 	
ï	 	Pandas (pd): library used for data manipulation and analysis. This library provides data 
structures  like  Data  Frame  and  Series,  and  functions  for  data  cleaning,  filtering,  and 
analysis. 	
ï	 	Numpy (np): library used for numerical computing in python. It offers support for large, 
multi-dimensional arrays and matrices, along with mathematical functions. 	
ï	 	Matplotlib.pyplot  (plt):  A  plotting  library  used  for  creating  static,  animated  and 
interactive visualizations. Commonly used for creating charts and plots. 	
ï	 	Statsmodels: the library used for statistical modelling and analysis. 	
ï	 	Seaborn  (sns):  is  a  data  visualization  library  based  on  Matplotlib,  and  it  provides  a 
high-level interface for creating attractive and informative statistical graphics. 	
ï	 	Sklearn.metrics:  This  module in  scikit-learn  (sklearn)  provides  a variety  of  metrics  for 
evaluating  the  performance of  machine learning  models.  These  metrics  cover  different 
aspects of model performance, depending on the type of task. 	
ï	 	Scipy:  is  a  scientific  library  in  Python  that  builds  on  the  capabilities  of  NumPy  and 
provides additional functionality for scientific computing. 	
7. Final output of the project 	
This  study  was  undertaken  to  analyses  the  seasonal  variations  in  the  farm  price  of 	
coconut  in  Kerala  and  to  develop  a  model  that  predicts  the  future  farm  prices  of  coconut. 
Here  SARIMA  (2,  1,  10,  12)  is  the  best  model  for  this  data  and  that  predicts  the  monthly 
average  farm  price  of  coconut  in  Kerala  for  the  next  1  year.  The  values  seem  to  show  an 
increasing trend over the months, with some fluctuations. This study helped to forecast the 
future price of coconut based on historical data at reasonable accuracy (MAPE=1.98). The 
results  indicated  that  the  SARIMA  based  farm  price  prediction  model  has  proven  to  be  a 
valuable tool for forecasting agricultural prices in Kerala.  	
The forecasted prices align closely with observed values, indicating that the model has 	
successfully  learned  from  historical  data  and  is  capable  of  making  reliable  predictions.	 	
While  the  model  has  shown  satisfactory  performance,  continuous  refinement  and  updates 
may further enhance its accuracy. Future iterations could consider incorporating additional 
factors  that  might  influence  farm  prices.	 Further  research  and  iterative  improvements  will 	
contribute to the sustained effectiveness of such forecasting models. 
8. References used  	
i. Price Statistics from 2000 to 2020 from Department of Economics & Statistics 	
ii.	 Compendium of Project Reports #AIFORALL Capacity Building in Artificial 	
Intelligence & Data Analytics from Department of Economics & Statistics 	
iii.	 Official website of Department of Economics & Statistics  	
iv.	 Agriculture Statistics at a glance 2022 from Ministry of Agriculture and Farmersâ 	
Welfare, Govt. of India.

40

41 	
 	
 	
 
 
 
 
 	
 
 
 	
Forecasting Paddy Yield  	
Using Sample Check Data 	
 	
 
 
 	
Submitted by 	
Kum. Lakshmi S	,  	
Statistical Assistant Grade II

42 	
 	
 	
Abstract 
This  study  utilizes  sample  check  survey  data  on  paddy  cultivation  to  develop  an  Artificial 
Neural  Network  (ANN)  model  focused  on  predicting  yield  weight.  The  dataset  includes 
various variables such as district, manuring status, irrigation, fertilizer quantities (N, P, K), 
seed  variety,  and  yield  weight.  The  primary  objective  is  to  leverage  this  data  to  construct 
an  ANN  model  that  accurately  predicts  paddy  yield  weight.  The  model  aims  to  offer 
valuable  insights  into  the  factors  influencing  yield,  contributing  to  the  advancement  of 
predictive models in agricultural contexts. 
 
Introduction. 
The program for sample check on area enumeration involves the identification and location 
of  four  clusters  within  each  selected  Investigator  zone.  Differences  in  crop  reporting  and 
recording  across  survey  numbers  during  various  seasons  are  pinpointed  by  supervisory 
officers from NSO or DES through comprehensive analysis. 
Concurrently,  the  inspection  of  crop  cutting  experiments  during  the  harvest  stage  entails 
evaluating  the  adherence  of  investigators  to  prescribed  General  Crop  Estimation  Survey 
procedures. This assessment encompasses multiple facets, including the selection of fields 
and  random  coordinates,  marking  of  experimental  plots  with  specified  dimensions, 
harvesting  procedures,  weighing  of  the  produce,  among  others.  Additionally,  the 
examination encompasses scrutinizing the provision and utilization of equipment for these 
experiments,  the  training  received  by  primary  workers,  crop  conditions,  input  usage,  and 
related factors. 
Utilizing  the  data  acquired  from  this  survey,  I've  gathered  comprehensive  information  on 
paddy cultivation, encompassing district-specific details, seed varieties, irrigation methods, 
manure application status, and specific quantities of N, P, K fertilizers used. Additionally, 
the dataset includes the recorded weights of the paddy yield. 
 	
Machine  learning  models,  known  for  their  versatility  and  predictive  capabilities, 	
find  extensive  applications  across  various  industries.  These  models,  designed  to  analyze 
patterns and make predictions from data, have a significant potential for application within 
the  agricultural  sector.  For  instance,  in  agriculture,  these  models  can  be  employed  to 
predict  crucial  outcomes  such  as  crop  yield.  By  leveraging  data  encompassing  diverse 
variables  like  district-specific  details,  seed  varieties,  irrigation  methods,  manure 
application status, and specific quantities of fertilizers (N, P, K), machine learning models 
can  effectively  forecast  paddy  yield  weight.  This  predictive  ability  enables  farmers  and 
agricultural  experts  to  anticipate  yield  trends,  optimize  resource  allocation,  and  make 
informed  decisions  to  enhance  crop  productivity.  Moreover,  machine  learning  models  can 
aid  in  disease  prediction  in  crops,  optimize  irrigation  schedules,  suggest  suitable  crop 
varieties  based  on  environmental  conditions,  and  even  predict  market  demand,  thereby 
assisting  farmers  in  strategic  planning  and  maximizing  agricultural  output.  The  utilization 
of  machine  learning  models  in  agriculture  presents  a  promising  prospect  for 
revolutionizing farming practices, improving crop yields, and ensuring global food security 
by enabling data-driven decision-making in this vital sector.

43 	
 
 	
 	
Objective 
To  construct  an  Artificial  Neural  Network  (ANN)  model  using  paddy  sample  data, 
comprising  district,  whether  manure  was  applied  or  not,  irrigation  status,  fertilizer 
quantities  (N,  P,  K),  seed  variety,  and  yield  weight.  The  objective  is  to  accurately  predict 
the yield weight based on these variables. 
 
 	
 	
Distribution of weight of yield according to Seed Variety. 	
 
 	
          High Yield 	
Local

44 	
 	
 	
Distribution of weight of yield according to Irrigation 	
 	
 	
 
  Unirrigated  Irrigated

45 	
 	
 
 	
Distribution of Weight of Yield according to Manure 	
 	
 	
 
 
  Not manured                  Manured

46 	
 	
 
 	
Visualization of Fertilizer composition impact on Paddy Yield 	
 	
Maximum weight was obtained for 	N	 :117,  	P:60, 	K	:74 	
 	
Combined influence of Manure, Irrigation and Seed variety on Paddy yield weight 	
 	
 	
 	
 	
Methodology 	
 	
1. Data Collection and Preprocessing 	
ï· 	Data Sources 	
The  data  used  in  this  project  was  gathered  from  various  zones  across  14  districts  within 
Kerala state. It comprised 1198 observations, encompassing diverse parameters relevant to 
paddy yield prediction.

47 	
 
 
 	
ï· 	Data Preprocessing 	
Conversion  of  Categorical  Data:  District  names,  a  categorical  variable,  were  transformed 
into numerical representations (1-14) to enable their use in predictive modeling. 
 
Yield  Weight  Classification:  The  weight  of  paddy  yield  was  classified  into  six  distinct 
categories  (1-5  kg,  5-10  kg,  10-15  kg,  15-20  kg,  20-25  kg,  and  25-30  kg).  This 
categorization enabled the modelling of yield ranges. 
2. Model Development using Artificial Neural Networks (ANN	) 	
 	
ï· 	Feature Engineering 	
Normalization/Scaling:  Input  features  underwent  normalization  or  scaling  to  ensure 
uniformity in their ranges and prevent bias during model training. 
 	
ï· 	Categorical Variable Encoding 	
Techniques  such  as  one-hot  encoding  or  label  encoding  were  employed  to  handle 
categorical  variables  (e.g.,  irrigation,  manure  status,  seed  variety)  for  compatibility  with 
the neural network. 
 	
ï· 	Model Architecture 	
 
Neural  Network  Configuration:  An  Artificial  Neural  Network  (ANN)  architecture  was 
constructed, comprising layers: 
 
Input Layer	: Configured to accommodate the number of input features. 	
Hidden  Layers:  Comprised  64  and  32  neurons  respectively,  employing  'relu'  activation 
functions. 
Output  Layer	:  Consisted  of  7  neurons  employing  'softmax'  activation  for  multi-class 	
classification, predicting yield range categories. 
 	
ï· 	Model Training and Evaluation 
 	
Data  Splitting	:  The  dataset  was  divided  into  training  (80%)  and  testing  (20%)  sets  for 	
training and validating the model's performance. 
 
Model  Compilation	:  The  ANN  was  compiled  using  'categorical_crossentropy'  as  the  loss 	
function and the 'adam' optimizer. 
 
Training Process	: The model was trained over 64 epochs with a batch size of 32, and a 10% 	
validation split was used to monitor and mitigate overfitting during training. 
 
Model  Accuracy	:  The  trained  model  achieved  an  accuracy  of  87%  on  the  test  dataset, 	
indicating its ability to predict paddy yield ranges effectively.

48 	
 
 	
ï· 	Model Improvement Strategies To enhance the model's performance: 	
 
Hyperparameter  Tuning	:  Experimentation  was  conducted  with  different  epochs,  batch 	
sizes, and learning rates. 
Regularization Techniques	: Techniques like dropout were explored to prevent overfitting. 	
Feature  Engineering	:  Additional  feature  engineering  and  selection  methods  were 	
considered to improve the model's predictive power. 
 
 
3. Result Analysis and Conclusion 
 
The  methodology  section  concludes  with  an  analysis  of  the  model's  predictions  and 
performance metrics, such as accuracy, precision, recall, and F1-score. The iterative nature 
of  refining  the  model  based  on  dataset  understanding  and  achieving  an  accuracy  of  87% 
was  highlighted.  Additionally,  potential  avenues  for  future  enhancements  or  alternative 
modelling approaches were proposed. 
 	
Prediction of Unseen Data 	
Based  on  the  model's  prediction,  for  the  given  conditions  in  the  Autumn  season  in 
Pathanamthitta  district,  employing  irrigation,  using  N:P:K  fertilizers  in  the  proportion  of 
15,20,15  respectively,  without  manure,  and  planting  local  seed  variety,  the  model  predicts 
a yield between 5 and 10 kilograms of paddy. 
This incorporation of a specific prediction for an unseen scenario not present in the training 
dataset demonstrates the practicality and usefulness of the model in making predictions for 
real-world agricultural conditions. 
 
Reference	 	
 
Journal Articles: 
1.Thomas,  R.,  &  Bhat,  R.  (2018).  Predictive  Modeling  of  Crop  Yield  using  Machine 
Learning Techniques. International Journal of Computer Applications, 180(15), 32-39. 
 
2.Das,  S.,  Mishra,  P.,  &  Mohanty,  S.  P.  (2020).  Impact  of  Irrigation  and  Fertilizer 
Application on Crop Yield: A Case Study in Eastern India. Journal of Agricultural Science 
and Technology, 22(5), 1101-1115. 
 
Online Sources: 
1.United  States  Department  of  Agriculture  (USDA):  National  Agricultural  Statistics 
Service. (n.d.). https://www.nass.usda.gov/ 
2.Agricultural  Research  Service  (ARS):  United  States  Department  of  Agriculture.  (n.d.). 
https://www.ars.usda.gov/ 
3.International Food Policy Research Institute (IFPRI).  
https://www.ifpri.org/ 
4.Global Open Data for Agriculture and Nutrition (GODAN).  https://www.godan.info/ 
5.Food  and  Agriculture  Organization  (FAO)  Statistical  Databases.  
http://www.fao.org/faostat/en/#data 
6.AgriTech Hub: Agricultural Technology Resources. https://www.agritechhub.com/

49 	
 
 
 	
 	
 
 
 
 
 	
Analysis and Forecasting of 	
Tapioca Production in Kerala 	
 	
 	
 
 
 
 	
Submitted by 	
Kum. Reshmi S,  	
Statistical Assistant Grade II

50 	
 
Introduction 	
 	
In the lush landscapes of Kerala, India, where agriculture plays a pivotal role in the 	
economy, tapioca production stands as a significant contributor to the agricultural tapestry. 
The  subtropical  climate  and  fertile  soil  of  Kerala  provide  an  ideal  environment  for 
cultivating  tapioca,  a  starchy  root  crop  widely  utilized  in  various  culinary  traditions.  This 
analysis  delves  into  forecasting  tapioca  production,  leveraging  ARIMA  modelling 
techniques.  Recognizing  the  importance  of  accurate  production  forecasts  for  informed 
agricultural  planning,  the  study  aims  to  unravel  patterns  and  trends  in  historical  data.  By 
examining  the  intricacies  of  tapioca  cultivation  in  Kerala,  this  analysis  not  only  seeks  to 
enhance predictive modelling but also to offer valuable insights into the factors influencing 
the  region's  tapioca  production.  The  unique  agro-ecological  conditions  of  Kerala,  coupled 
with  the  prominence  of  tapioca  in  its  agricultural  landscape,  make  this  study  particularly 
relevant  for  stakeholders  in  the  region,  providing  a  foundation  for  informed  decision-
making and sustainable agricultural practices. 
Objectives 
The  primary  objective  of  this  report  is  to  employ  ARIMA  modeling  techniques  for 
forecasting  tapioca  production  in  Kerala,  India.  By  leveraging  time  series  analysis,  the 
study  aims  to  discern  patterns,  trends,  and  potential  influencing  factors  in  historical  data 
related  to  tapioca  cultivation  in  the  region.  The  specific  objectives  include  identifying 
optimal  orders  (p,  d,  and  q)  for  the  ARIMA  model,  generating  accurate  forecasts,  and 
evaluating  the  performance  of  the  forecasting  model.  Through  this  analysis,  the  report 
strives  to  provide  valuable  insights  into  the  temporal  dynamics  of  tapioca  production  in 
Kerala,  facilitating  informed  decision-making  for  agricultural  planning  and  resource 
management. 
Datasets Used 
To  conduct  this  analysis  of  tapioca  production  in  Kerala,  i  used  a  comprehensive  dataset 
that  contains  relevant  informations  about  area  of  cultivation  of  tapioca,  annual  tapioca 
production, annual rainfall of 37 years from 1985 to 2022.

51 	
 
Data Source	: 	
       Dataset  is  collected  from  our  publication  â  âAgricultural  Statistics,  Department  of 
Economics and Statistics, Govt. Of Keralaâ. 
Methods used for analyse the relationship between the factors such as area and rainfall. 
 
Tools and Libraries used 
Python is  used for the analyse the correlation and time series analysis of the data.  For this 
python  and  google  collab  notebook  is  used.  The  data  analysis  is  done  using  python 
libraries such as Matplotlib, Pandas, Numpy etc.The ARIMA model for prediction is done 
using R-Programming. Statsmodels libraries are used. 
 	
Visualisation of the correlation using Heat Map 	
 	
 	
 
Correlation Matrix	 	
   Area total         Tapioca Production        Rainfall 	
 
Area total	                         1.000000              0.688104                 -0.046007 	
Tapioca Production	         0.688104              1.000000                 -0.082406 	
Rainfall 	                           -0.046007            -0.082406                 1.000000 	
Area total has positive relation. Rainfall shows negative relation.

52 	
 
Graphical representation of Tapioca production over time	 	
 	
Auto Regressive Integrated Moving Average [ARIMA Model] 
      An  ARIMA  is  a  statistical  analysis  model  that  uses  time  series  data  to  either  better 
understand  the  dataset  to  predict  the  future  trends.  It  predicts  the  future  values  based  on 
past values. 
Autocorrelation  Functions  (ACF)  and  Partial  Autocorrelation  Functions  (PACF)  plots  are 
useful  tools  in  time  series  analysis  to  identify  the  autocorrelation  structure  of  the  data. 
From these graph we can understand the relationship between observations at different lags.  
 
ACF and PACF

53 	
 	
 	
 
 
 
gmented Dickey-Fuller Test 
 
Dickey-Fuller = -4.6098 
Lag order = 3 
p-value = 0.01 
Alternative hypothesis: stationary 
 
Considering different models using Auto ARIMA 
 
ARIMA(2,2,2)                    : 951.7914 
 ARIMA(0,2,0)                    : 986.9408 
 ARIMA(1,2,0)                    : 981.0933 
 ARIMA(0,2,1)                    : Inf 
 ARIMA(1,2,2)                    : Inf 
 ARIMA(2,2,1)                    : 957.0147 
 ARIMA(3,2,2)                    : 952.5215 
 ARIMA(2,2,3)                    : Inf 
 ARIMA(1,2,1)                    : 961.7851 
 ARIMA(1,2,3)                    : 955.9423 
 ARIMA(3,2,1)                    : 954.6358 
 ARIMA(3,2,3)          Inf 
 
Best model: ARIMA(2,2,2)

54 	
 
  
  
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Result

55 	
 
Forecasted production  for next 3 years(2023-2025)	 	
 
           	Point Forecast           Lo 95          Hi 95	 	
2022         2590419            2257713    2923125 
2023         2882707            2437173    3328241 
2024         2578168            2102395    3053941 
 
               ME        RMSE      MAE       MPE     MAPE      MASE     	Training set	 25396.12 	
 155375.2  116605.2  0.8623782 4.412618  0.7440292  
 
ACF1 
 
0.022996 
 
 
Conclusion 
In  conclusion,  this  project  aimed  to  enhance  our  understanding  of  tapioca  production  in 
Kerala through the application of ARIMA modeling techniques.  By leveraging time series 
analysis  on  historical  data,  we  successfully  identified  optimal  orders  (p,  d,  q)  for  the 
ARIMA  model,  enabling  us  to  generate  accurate  forecasts  for  tapioca  production.  The 
results, validated through key metrics such as Mean Absolute Error (MAE), Mean Squared 
Error (MSE), Root Mean Squared Error (RMSE), and R-squared, provided insights into the 
temporal dynamics of tapioca cultivation in the region. 
 
The analysis revealed patterns and trends, shedding light on the factors influencing tapioca 
production  in  Kerala.  The  unique  agro-ecological  conditions,  climate  variations,  and 
potentially  other  contributing  variables  were  considered  in  interpreting  the  forecasting 
results.  While  acknowledging  the  challenges  associated  with  forecasting  agricultural 
commodities,  the  findings  of  this  study  contribute  to  the  ongoing  efforts  in  sustainable 
agricultural planning and resource management in Kerala. 
This  project  not  only  deepens  our  knowledge  of  tapioca  production  but  also  underscores 
the  importance  of  accurate  forecasting  for  informed  decision-making  in  agriculture.  The 
insights  gained  can  guide  policymakers,  farmers,  and  other  stakeholders  in  optimizing 
resource allocation, improving crop yield predictions, and fostering resilience in the face of 
changing  agricultural  landscapes.  Looking  ahead,  the  lessons  learned  from  this  study  can 
serve as a foundation for further research and refinement of forecasting models.

56

57 	
 	
 	
 
 
 
 
 
 
 	
Consumer Price Index (R/U/C) 	
A Machine Learning Approach 	
to Cluster the Markets 	
 	
 	
 
 	
Submitted by 	
Sri. Sajin Gopi,  
Assistant Director

58 	
 	
1. Introduction 	 	
The  Consumer  Price  Index  (CPI)  stands  out  as  a  key  and  widely  utilized  economic 
indicator  for  assessing  the  cost  of  living  for  specific  segments  of  the  population  at  a 
given  point  in  time.  As  implied  by  its  name,  the  CPI  is  intricately  linked  to  the  price 
fluctuations of commodities and the purchasing behaviour of the population, with these 
aspects  often  serving  as  reflections  of other  economic parameters.  The  Consumer Price 
Index  (Rural/Urban/Combined)  holds  key  importance  as  it  represents  all  strata  of  the 
population,  including  the  Poor,  Middle,  and  Affluent.  The  dynamics  of  the  market, 
whether  Rural  or  Urban,  play  a  pivotal  role  in  shaping  the  value  of  CPI(R/U/C)  since 
rural  and  urban  markets  may  exhibit  significant  variations  in  the  aforementioned  price 
fluctuations  and  purchasing  patterns.  The  objective  of  the  study  is  to  assess  the 
significance  of  classifying  markets  into  Rural  and  Urban  categories  in  Kerala  for  the 
generation of CPI(R/U/C). 
 
2.Scope and Structure of the problem 
The  Laspeyres  Formula  in  formulating  CPI(R/U/C),  relying  on  the  weights  assigned  to 
different  commodities  grouped  into  six  primary  Groups  and  their  corresponding 
Subgroups. These weight assignments are derived from the outcomes of the 68th round 
of the National Sample  Survey's Consumer Expenditure Survey (CES). There  are  total 
of  149  markets  across  all  districts,  comprising  78  urban  and  71  rural  markets,  where 
prices  of  these  commodities  are  regularly  collected.  The  pricing  of  each  commodity  is 
subject to common factors such as local availability, transportation, and demand, which 
are  considered  usual  in  nature.  As  highlighted  earlier,  the  prices  of  approximately  four 
hundred  items  categorized  into  groups  and  subgroups  may  exhibit  variations  between 
rural  and  urban  areas.  The  approach  involves  clustering  markets  based  on  these  price 
variations. 	
Since  Kerala  is  a  consumer  state,  it  heavily  relies  on  neighbouring  states  for 	
almost  all  essential  commodities.  The  lifestyle,  occupations  of  the  people,  and  various 
development  indicators  in  the  state  suggest  that  it  is  undergoing  rapid  urbanization, 
distinguishing it as one of the states in the country experiencing substantial urbanization. 
Here  comes  the  attention  to  a  machine  learning  approach  to  cluster  the  markets  in 
Kerala to enable the impact of Rural and Urban classification on CPI.  
 	
The  objective  is  employing  a  machine  learning  approach  to  cluster  markets  in 	
Kerala based on the influence of Rural and Urban classifications on the Consumer Price 
Index  CPI(R/U/C).  This  initiative  holds  the  potential  to  offer  valuable  insights  into  the 
distinctions  between  Rural  and  Urban  markets  considering  the  price  values  of 
commodities.  Key  steps  include  identifying  relevant  features,  pre-processing  the  data, 
selecting a suitable clustering algorithm, feature engineering, model training, evaluation, 
interpretation of results, visualization, and a detailed analysis of the impact of Rural and 
Urban  classifications  on  the  formed  clusters.  The  ultimate  goal  is  to  provide  a 
comprehensive understanding of how CPI(R/U/C) item basket price values contribute to 
market  clustering  and  the  subsequent  differentiation  between  Rural  and  Urban 
classifications.

59 	
 
3. K-Means Clustering 	
Clustering  is  an  unsupervised  learning  technique  designed  to  partition  data  into 	
distinct  groups  where  the  elements  within  each  group  exhibit  similarities.  The  primary 
objective  of  clustering  is  to  unveil  meaningful  and  significant  patterns  within  the  data. 
These  groups  can  serve  various  purposes,  such  as  direct  analysis,  in-depth  exploration,  or 
utilization  as  features  or  outcomes  for  predictive  regression  or  classification  models.  K-
means,  being  the  first  clustering  method  developed,  remains  widely  used  due  to  its 
algorithmic simplicity and effectiveness in handling large datasets. 	
In  our  scenario,  the  application  of  the  K-Means  algorithm  to  cluster  markets  into 	
two  groups  could  yield  satisfactory  model  accuracy  measures.  Comparing  these  clusters 
with  the  actual  market  labels  of  Rural  and  Urban  may  reveal  relevant  insights.  If  the 
algorithm successfully  generates two clusters resembling the actual rural and urban labels, 
it implies a distinct separation between markets in rural and urban areas. Essentially, in the 
context  of  Kerala,  there  exists  a  significant  divergence  between  markets  situated  in  rural 
and urban areas. 
4. Data Pre processing           	
The  monthly  market  data,  initially  unstructured,  underwent  a  three-stage  data  pre-	
processing  to  facilitate  model  building.  With  approximately  400  commodities  featuring 
different  codes  related  to  CPI  (R/U/C)  and  additional  characteristics,  the  goal  was  to 
extract  commodities  sharing  similar  attributes  across  markets,  creating  a  robust  dataset. 
This  process  involved  the  development  of  Python  code  comprising  three  user-defined 
functions: 	
 
 
1. Phase 1 Function: 	
-  To  extract  each  commodity  with  a  unique  item  code  from  the  Excel  file. 	
Included  features  such  as  unit,  observed  quantity,  base  price,  and  monthly  prices. 
The  output  was  a  dataset  encompassing  298  commodities  and  their  corresponding 
price values for a span of 6 months, from January 2020 to June 2020. 
2. Phase 2 Function: 	
-To Unify units (e.g., Kilogram, Numbers, Packets) based on a Unit Master. 	
Updated unit information consistently across markets for the dataset generated in 
Phase 1. 	
3. Phase 3 Function: 	
-  To  standardize  price  values  of  commodities  according  to  the  unique  unit 	
and  observed  quantity.  This  standardization  enabled  meaningful  market-to-market 
comparisons. 
 
These  three  functions  significantly  reduced  the  workload  in  pre-processing  data 	
for  149  markets.  The  efforts  resulted  in  successfully  obtaining  processed  data  for  139 
markets,  while  the  remaining  markets  exhibited  substantial  inconsistencies  that  proved 
challenging to pre-process. The resultant dataset showcases a structured and standardized 
representation  of  market  data,  facilitating  efficient  comparisons  across  139  markets. 
Upon further examination of the data, it became evident that the price values of the initial 
298  commodities  were  significantly  influenced  by  the  diverse  selection  of  products

60 	
 
during  the  item  basket  fixation  process,  leading  to  substantial  variations  across  markets. 
Recognizing the potential challenges posed by  the highly  inconsistent dataset, a decision 
was made to narrow the scope of the dataset to focus exclusively on six varieties of rice. 
Rice,  being  a  fundamental  commodity,  undergoes  close  monitoring  of  its  price 
fluctuations  across  markets.  Further,  uniformity  and  comparability  of  attributes  such  as 
unit,  observed  quantity,  and  base  price  across  all  markets  leads  to  streamline  the  dataset 
to include only rice varieties. This deliberate restriction enhances the dataset's  coherence 
with the study's objectives, ensuring a more meaningful and reliable foundation for model 
building. The data set looked like 	
 	
              	
 	
 
5. Tools an Packages used 
The  tools  and  libraries  utilized  to  construct  the  model  in  Python  include  the  following 
libraries: 
 
- `numpy` 
- `pandas` 
- `matplotlib` 
- `seaborn` 
- `sklearn` 	
These  libraries  were  instrumental  in  various  stages  of  the  model  development 	
process,  providing  essential  functionalities  for  data  manipulation,  visualization,  and 
machine learning tasks.

61 	
 
 
 
 
 
 
 
6. Graphical analysis of distribution of rice varieties over Urban and Rural markets	 	
 	
Figure1	 	
Figure 2

62 	
 	
Figure 3 	
 
The graphical analysis explains the distribution of six rice varieties across Rural and Urban 
markets.  Notably,  all  varieties  exhibit  a  nearly  identical  pattern,  leaning  slightly  towards 
the right tail. Both box plots and violin plots validate this observation. It is evident that the 
behaviour  of  all  six  varieties  remains  consistent  across  both  Rural  and  Urban  categories. 
While  certain  varieties  may  demonstrate  higher  distribution  values  in  Urban  markets 
compared  to  Rural  ones,  such  differences  are  insufficient  to  conclude  that  Urban  markets 
are generally more expensive than their rural counterparts. 
 
7. Model Building 
  K-means  algorithm  precisely  applied  to  cluster  the  all  834  data  points  two  cluster  with 
their under laying similarities. The below given plot exhibit the two clusters clearly. 
 	
18  366 
97  353

63 	
 	
 	
Figure 4 	
The  figure  visually  indicates  the  absence  of  a  definitive  classifier  or  boundary  to 
exclusively  distinguish  between  the  two  clusters,  despite  their  formation.  The  silhouette 
score, measuring the cohesion and separation of clusters, is only 0.482. This score suggests 
that  the  model  does  not  demonstrate  a  significant  difference  between  the  two  obtained 
clusters,  emphasizing  the  challenge  of  establishing  a  clear  distinction  based  on  the 
available features. The provided confusion matrix is,  
 
     
 
This  implies  that  out  of  the  total  834  data  points,  only  371  are  correctly  assigned  to  their 
respective  labels,  Rural  or  Urban.  The  precision  values  for  each  cluster,  at  16%  and  48%, 
further  underscore  the  model's  subpar  performance.  These  precision  values  indicate  the 
proportion of correctly identified instances within each cluster, highlighting the challenges 
in accurately classifying data points into their correct categories. 
The model's poor performance in clustering the dataset into two categories underscores the 
challenge  of  determining  the  actual  number  of  clusters  that  can  effectively  subdivide  the 
dataset.  The  Elbow  method  is  employed  to  address  this  issue.  This  technique,  commonly 
used in data analysis and machine learning, aims to identify the optimal number of clusters 
for  k-means  clustering.  The  method  entails  executing  the  k-means  clustering  algorithm 
across  a  range  of  k  values  (number  of  clusters)  and  plotting  the  sum  of  squared  distances 
from  each  point  to  its  assigned  centre  against  the  number  of  clusters.  The  "elbow"  in  the 
plot  signifies  a  point  where  the  rate  of  decrease  in  the  sum  of  squared  distances  slows 
down,  forming  a  bend  resembling  an  elbow.  The  objective  is  to  select  the  k  value  at  the 
elbow  point  because,  beyond  that  point,  additional  clusters  do  not  significantly  reduce the 
sum  of  squared  distances,  and  extra  clusters  may  not  yield  meaningful  separation  in  the 
data.

64 	
 
 
  Here is the elbow plot  	
 	
Figure 5 	
In  the  provided  elbow  plot  (Figure  5),  it  is  evident  that  no  distinct  elbow  point  is  present. 
This  absence  suggests  that  there  is  no  optimal  number  for  subdividing  the  dataset  into 
clusters.  In  other  words,  the  subgroups  within  the  dataset  exhibit  significant  similarity  in 
nature, making it challenging to identify a meaningful number of clusters.   For  instance 
the  attempt  to  cluster  the  data  set  into  five  clusters  gave    poor  clusters  as  shown  in  figure 
below. 	
 	
Figure 6 	
The figure visually demonstrates that the clusters formed exhibit overlap, indicating a lack 
of reliability in the clustering results.

65 	
 
8. Inference and Conclusion 	
The objective was to cluster markets in Kerala based on the price values of six rice 	
varieties.  The  outcomes  of  the  K-Means  clustering  analysis  imply  that  there  is  no 
significant difference in terms of price values between rural and urban markets. Simply put, 
the  prices  in  rural  and  urban  areas  do  not  vary  significantly.  This  observation  aligns  with 
the  earlier  mentioned  point,  supporting  the  inference  that  rural  areas  in  Kerala  are 
undergoing a substantial rate of urbanization. 
9. Opportunities for Further Investigation 	
Expanding  the  model  to  include  a  broader  range  of  commodities  over  an  extended 	
time  frame  could  yield  valuable  insights  for  market  selection,  replacement,  and  other 
analyses  related  to  price  values  in  the  computation  of  Consumer  Price  Index  (CPI)  for 
Rural,  Urban,  and  Combined  areas.  This  enhanced  scope  may  provide  a  more 
comprehensive  understanding  of  market  dynamics,  facilitating  more  informed  decision-
making in terms of commodity trends, market fluctuations, and economic scrutiny 
10. References 	
â 	Peter  Bruce,  Andrew  Bruce  and  Peter  Gedeck.  2020.  Practical  Statistics  for  Data 
Scientists, OâREILLY, Second edition 283-302. 	
â 	Methodology  for  Compilation  of  Consumer  Price  Index 
Numbers(Rural/Urban/Combined)  at  District  Level  In  Kerala,  Price  Division, 
Directorate of Economics and Statistics, Kerala 	
 
11. Appendix 
  Code for Phase1, Phase2, Phase3 functions written for data processing 	
1.Phse1 Function

66 	
 	
 	
 	
2.Phse2 Function

67 	
 	
3.Phse3 Function

68

69

70

71 	
 	
 	
 
 
 
 	
 
 
 
 	
Prediction of Farm Wholesale Price of Paddy 	
Cost of Cultivation as Predictor 	
 	
 
 
 	
Submitted by 
Sri. Saju K,  	
Research Assistant

72 	
 
1. Executive Summary 	
The  project  aims  to  develop  predictive  models  for  forecasting  the  wholesale  price 	
of paddy with cost of cultivation of paddy  as predictor. Two machine learning algorithms, 
Decision  Tree  Regression  and  Random  Forest,  are  employed  for  prediction.  The  project 
seeks  to  provide  farmers  and  stakeholders  with  valuable  insights  into  future  paddy  prices, 
enabling informed decision-making. 
2. Introduction 
2.1 Background 	
Agriculture  plays  a  crucial  role  in  the  state's  economy,  providing  livelihoods  to  a 	
significant  portion  of  the  population.  Paddy  cultivation  is  a  cornerstone  of  agriculture  in 
Kerala,  contributing  substantially  to  the  state's  food  production.  The  success  of  paddy 
farming directly impacts  the food security and economic well-being of the state. The farm 
price of paddy is a critical factor for farmers, influencing their income and overall financial 
stability.  Fluctuations  in  paddy  prices  can  significantly  affect  the  livelihoods  of  farmers, 
making accurate price prediction essential. The need for an efficient and reliable system for 
predicting  farm  wholesale  prices  of  paddy  in  Kerala  is  evident.  Agricultural  markets  are 
dynamic,  and  predicting  the  prices  of  crops  is  essential  for  farmers  and  traders  to  make 
strategic  decisions.  This  project  focuses  on  predicting  the  wholesale  price  of  paddy  using 
machine learning models. 
2.2 Objectives 	
ï· 	Develop a Decision Tree Regression model for predicting paddy prices. 	
ï· 	Implement a Random Forest model for comparison. 	
ï· 	Evaluate and compare the performance of both models. 	
3. Data Collection and Pre-processing 
3.1 	Data Source 	
The dataset used for this project is collected from the annual publications from two 	
sections namely Cost of cultivation and Prices of the Department of Economics and 
Statistics directorate. 
3.2 Data Pre-processing	 	
The  dataset  available  in  the  prices  section  is  in  the  form  of  monthly  data  while  the 	
cost  of  cultivation  is  in  the  form  of  season  wise  data  (Autumn,  Winter,  Summer).  So  the 
first step of the data pre-processing is to convert the monthly data in to season wise data by 
taking  the  average  of  four  months.  That  is  taking  average of  July,  August,  September and 
October to get the autumn season Farm price. After the pre-processing 23 years of 3 season 
data from  2000  to  2022  is  available for  the  project.  The dataset  used  and  visualization  are 
given below.

73 	
 	
Sl.No	 	cost_of_cultivation	 	price 	
1  16218  688.25	 	
2  17916  680.50	 	
3  19354  678.25	 	
4  17210  666.50	 	
5  16318  654.00	 	
6  20270  601.25	 	
7  20484  584.00	 	
8  18614  576.82	 	
9  21754  576.02	 	
10  19777  636.26	 	
11  18715  658.23	 	
12  20111  657.43	 	
13  19932  683.81	 	
14  18925  688.09	 	
15  19694  711.13	 	
16  18835  702.86	 	
.  .  . 
.  .  . 
67  75430  2038.41	 	
68  82025  2322.86	 	
69  80802  2302.13	 	
 	
Cost_of_Cultivation   -Rs/Ha	
 	
Price                         -Rs/Qtl

74 	
 	
 	
 	
 	
The  scatter  plot  shows  a  positive  correlation  between  cost  of  cultivation  and  farm 	
wholesale price of paddy.  Also the heatmap shows correlation coefficient of 0.98 between 
the  cost  of  cultivation  and  the  farm  wholesale  price  of  paddy  indicates  a  very  strong 
positive correlation.

75 	
 
4. Methodology 
4.1	  Tools and Libraries used 	
The predictive models were fitted using Python Programming. The techniques used 	
were Decision Tree Regression and Random Forest. The libraries used were    	
ï¶	 pandas 	
ï¶	 joblib	 	
ï¶	 sklearn	 	
ï¶	 matplotlib	 	
ï¶	 math	 	
4.2 Decision Tree Regression 	
Decision  Tree  Regression  is  employed  to  build  a  model  that  recursively  splits  the 	
dataset based on features, creating a tree-like structure. This model is suitable for capturing 
non-linear relationships in the data. 
Advantages:	 	
ï· 	Interpretability:	 Decision trees offer a clear and interpretable representation. 	
ï· 	Handling Non-linearity:	 Effective in capturing non-linear relationships in data. 	
ï· 	Feature Importance:	 Provides insights into the importance of different features. 	
4.3  Random Forest 	
Random  Forest  is  an  ensemble  learning  method  that  combines  multiple  Decision 	
Trees  to  improve predictive accuracy  and  control  over  fitting.  It  works  by  training  several 
trees on different subsets of the data and averaging their predictions. 
Advantages:	 	
ï· 	High Accuracy:	 Random Forest tends to provide accurate predictions. 	
ï· 	Reduced  Over  fitting:	  Randomization  helps  mitigate over  fitting  seen  in  individual 	
decision trees. 	
ï· 	Feature  Importance:	  It  can  assess  the  importance  of  different  features  in  making 	
predictions. 	
5. Model Training and Evaluation 
5.1 Decision Tree Regression Model 	
ï· 	Trained the model on the training dataset. (80% of the Dataset) 	
ï· 	Evaluated the model on the testing dataset. (20% of the Dataset) 	
ï· 	Measured performance using metrics such as Mean Squared Error (MSE), Mean 
Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared.

76 	
 
Observed error values are given below. 
Mean Squared Error          : 7782.09083714875 
Mean Absolute Error         : 56.074275071428566 
Root Mean Squared Error : 88.21615972795885 
R-squared                          : 0.9739485254884002 
Actual v/s Predicted Values 
 	A  graph  is  plotted  with  actual  and  predicted  values  of  the  Farm  wholesale  price  of 	
paddy.  By  analyzing  the  graph  we  can  see  that  the  predicted  values  are  approximately 
nearer to the actual values. 	
 	
Residual Plot 	
A residual plot is plotted to assess the goodness of fit of the fitted model. It helps to 	
identify patterns or trends in the residuals, which are the differences between the observed 
and  predicted  values  of  the  dependent  variable.  Analyzing  residual  plots  is  crucial  for 
understanding  whether  a  regression  model  captures  the  underlying  patterns  in  the  data 
adequately.

77 	
 
Random Prediction 	
A  random  prediction  is  done  using  the  value 	80,000	  as  cost  of  cultivation.  The 	
predicted value of the Farm wholesale price is 	2302.13375.	 	
5.2 Random Forest Model 	
ï· 	Train the Random Forest model on the training dataset. (80% of the Dataset) 	
ï· 	Evaluate the model on the testing dataset. (20% of the Dataset) 	
ï· 	Measured performance using metrics such as Mean Squared Error (MSE), Mean 
Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared. 	
Observed error values are given below. 
Mean Squared Error          : 9552.055205388411 
Mean Absolute Error         : 57.51485139500009 
Root Mean Squared Error : 97.73461620832411 
R-squared                          : 0.9680233592842841 
Actual v/s Predicted Values 	
By  analyzing  the  graph  we  can  see  that  the  predicted  values  are  approximately 	
nearer to the actual values. 	
 	
Residual Plot 	
Analyzing  the  residual  plot  residuals  are  randomly  scattered  around  zero,  with  no 	
discernible pattern, it indicates that the model's predictions are unbiased and capturing the 
underlying relationships in the data.

78 	
 	
 	
Random Prediction 	
A  random  prediction  is  done  using  the  value 	80,000	  as  cost  of  cultivation.  The 	
predicted value of the Farm wholesale price is 	2300.9263 	
 	
6. Comparison of the Results 	
A comparison of the results of Decision Tree Regression and Random Forest model 	
evaluations by the performance metrics are given below. 
 	
 
 	
MSE and RMSE:	 	
Decision  Tree  has  lower  MSE  and  RMSE  values,  indicating  better  performance  in 	
terms  of  average  squared  and  absolute  differences.  The  RMSE  of  the  Decision  Tree 
Regression  model  is  88.2,  which  is  higher.  But  comparing  the  Farm  wholesale  price  of 
paddy which is nearer to 2300, Rs.88 is a negotiable price. 
MAE:	 
Decision  Tree  also  has  a  lower  MAE,  suggesting  smaller  absolute  differences 	
between predicted and actual values. 
RÂ²:	 	
Decision  Tree  has  a  higher  RÂ²,  indicating  a  better  overall  fit  in  explaining  the 	
variance in the target variable.  	
Decision Tree Regression 	
Mean Squared Error         : 7782.09 
Mean Absolute Error        : 56.07 
Root Mean Squared Error: 88.21 
R-squared                          : 0.9739 	
Random Forest 	
Mean Squared Error         : 9552.05 
Mean Absolute Error        : 57.51 
Root Mean Squared Error: 97.73 
R-squared                          : 0.9680

79 	
 
 
7. Conclusion 	
After  comparing  the  performance  metrics  of  two  models  we  can  find  that  both 	
model  perform  approximately  the  same.  Even  though  Decision  Tree  Regression  model 
with 97.39% accuracy is more accurate than the Random Forest model. Based on the above 
values,  the  Decision  Tree  Regression  model  appears  to  outperform  the  Random  Forest 
model for the given dataset.  
8. References 
Publication  1. Report on Cost of cultivation of important crops in Kerala 2000 to 2022 	
2.Price Bulletin 2000 to 2022  
3.Regression model based studies 	
 	
This project report provides a comprehensive overview of the process of predicting 	
the wholesale price of paddy using Decision Tree Regression  and Random Forest models. 
It  encompasses  data  collection,  pre-processing,  model  development,  evaluation,  and 
comparison, offering valuable insights for stakeholders in the agriculture sector

80

81 	
 	
 	
 
 
 
 
 	
An Analysis of Suicides in Kerala  	
for the past 10 years 	
 	
 	
 
 	
Submitted by 	
Sri. Shibu B.T, 	Research Assistant	  	
and	 	
Sri. Brijesh C.J, 	Statistical Assistant Grade II

82 	
 	
Introduction 
 	
Suicide  is  a  multidimensional  and  serious  social  problem  that  affects  people  all 	
around  the  world,  regardless  of  where  they  live.  Kerala,  a  state  renowned  for  its  beautiful 
scenery  and  high  human  development  index,  is  not  exempt  from  the  alarmingly  high  rate 
of suicides. The goal of this research is to provide light on the underlying causes, patterns, 
and trends related to suicides that occurred in Kerala between 1996 and 2021. 	
Kerala  has  an  impressive  standard  of  living  (life  expectancy,  literacy,  health  care, 	
and  other  areas),  so  it  makes  sense  that  issues  of  socioeconomic  development  have 
received a lot of attention. However, there is another startling trend in Kerala that receives 
little  attention  from  social  science  research:  Kerala  has  the  fourth-highest  rate  of  suicides 
in  India.  It  is  crucial  to  take  a  close  look  at  the  less  talked-about  subject  of  mental  health 
and  wellbeing.  Suicide  needs  to  be  thoroughly  investigated  in  order  to  develop  effective 
preventive efforts because it is frequently believed that suicide is a symptom of underlying 
emotional, social, and economic difficulties. 	
According  to  the  publication  of  National  Crime  Records  Bureau  âAccidental 	
Deaths & Suicides in India 2022â A total of 1,70,924 suicides were reported in the country 
during  2022  showing  an  increase  of  4.2%  in  comparison  to  2021  and  the  rate  of  suicides 
has  increased  by  3.3%  during  2022  over  2021.  Number  of  suicides  reported  in  Kerala 
during  2022  is  10162  (5.9%  of  Total  suicides  in  India)  showing  an  increase  of  6.4%  in 
comparison to 2021. 	
Rate  of  suicides  i.e.  the  number  of  suicides  per  one  lakh  population,  has  been 	
widely accepted as a standard yardstick for comparison. All India rate of suicides was 12.4 
during the year 2022. Sikkim reported the highest rate of suicide (43.1) followed by A & N 
Islands  (42.8),  Puducherry  (29.7),  Kerala  (28.5)  and  Chhattisgarh  (28.2).  The  rate  of 
suicides  in  Kerala  for  the  years  2020,  2021  &  2022  are  24.0,  26.9  &  28.5  respectively. 
Notably, Kerala's suicide rate has exhibited an upward trend, increasing from 24.0 in 2020 
to  28.5  in  2022,  surpassing  the  national  averages  of  11.3,  12.0,  and  12.4  for  the  same 
respective  years.  Despite  its  smaller  size,  Kerala's  consistently  higher  rates  underscore  a 
significant concern for the state. 
Scope & Objectives of the study 	
The  scope  of  this  study  encompasses  a  comprehensive  examination  of  suicide 	
trends in Kerala from 1996 to 2021. The focus will extend to analysis the gender wise and 
age group wise causes of suicides.

83 	
 
 
Objectives 	
1.	 	Gender-wise Analysis:	 	
The  study  aims  to  examine  suicide  data  with  a  specific  focus  on  gender. 	
This  involves  analyzing  the  number  of  suicides  among  males  and  females 
separately.  It seeks to identify if there  are any significant differences in suicide 
numbers  between  genders,  and  if  so,  what  factors  may  contribute  to  these 
differences. 	
2.	 	Cause-wise Analysis:	 	
This  objective  involves  categorizing  and  studying  suicides  based  on  their 	
causes.  Causes  may  include  factors  such  as  health  issues,  socio-economic 
problems, family  conflicts, or other specific  reasons mentioned in the available 
data.  The  goal  is  to  understand  the  predominant  causes  behind  suicides  in 
Kerala and any variations in these causes over the years. 	
3.	 	Age Group-wise Analysis:	 	
The  study  will  examine  suicide  data  by  categorizing  it  into  different  age 	
groups.  This  objective  seeks  to  identify  whether  certain  age  groups  are  more 
vulnerable to suicide and to understand the patterns and trends in suicide across 
different  age  groups.  This  analysis  can  provide  insights  into  age-specific  risk 
factors. 	
4.	 	Year-wise Analysis:	 	
This  objective  involves  examining  the  suicide  data  on  a  yearly  basis.  The 	
study  aims  to  identify  any  temporal  trends,  fluctuations,  or  significant  changes 
in suicides over the years from 1996 to 2021. By analyzing the data on a yearly 
basis,  the  study  can  capture  the  dynamics  and  evolution  of  suicide  patterns  in 
Kerala. 	
In  summary,  the  study's  primary  focus  is  to  conduct  a  detailed  analysis  by 	
considering multiple dimensions such as gender, cause, age group, and year. This approach 
enables  a  comprehensive  understanding  of  the  various  factors  influencing  suicides  in 
Kerala, leading to insights that can inform targeted interventions, policy recommendations, 
and mental health initiatives tailored to specific demographics and contexts.

84 	
 
 
Limitations of the study 	
The study focused on suicide data in Kerala from 1996 to 2021, considering factors 	
such  as  sex,  age  groups,  and  causes.  Using  artificial  intelligence,  an  attempt  was  made  to 
predict  total  suicides  using  the  Auto  Regressive  Integrated  Moving  Average  (ARIMA) 
model. Due to the limited 26-year dataset, accuracy was  generally low in most AI models. 
Also  the  study  could  be  more  effective  with  the  inclusion  of  factors  like  profession, 
income, and educational qualification. 
Data set 	
Data  set  for  the  study  is  provided  by  the  Statistical  wing  of  State  Crime  Records 	
Bureau. The dataset contains the following columns. 	
Year :-	 This contain information from 1996 â 2021. 	
Cause:-	  This  column  tell  us  about  the  reason  of  committing  suicide  like  Family 	
Problems,  illness, Bankruptcy, unemployment etc.  	
Age  group  :	-  This  column  contains  the  cause  wise  total  suicide  committed  by  the 	
male & female age groups from 0-14 to 60+.  
Tools Used 	
Python Tools   	
â¢ 	 Pandas 	
â¢ 	 Numpy 	
â¢ 	Seaborn  	
â¢ 	SK Learn 	
â¢ 	Statsmodels 
Machine Language 	
â¢ 	ARIMA 	
Gender wise  Analysis 	
Among the total suicides, a significant gender skew is evident. A staggering 73.7% 	
of the reported suicides were males, totalling 1,68,562, while females accounted for 26.3%, 
with 60,004 cases.

85

86 	
 
Cause wise Analysis 
 
  Around 34% of the total suicides committed during the study period (1996-2021) is 
conducted due to family problem and next is due to illness (26%). While analysing male & 
female  suicides  separately,  it  is  noticed  that  34%  of  male  suicides  were  committed  due  to 
family problem and 25% is due to illness. In case of female it is 34% and 29% respectively.

87 	
 	
Cause wise Trends

88 	
 
Age Group-wise Analysis 
 	It is examined that the age group committing more suicides are 30-44 & 45-59. In 	
case of male it is 45-59 and in case of female it is 15-29.

89

90

91 	
 
 
Year-wise Analysis 	
  	 	While analysing the yearwise suicides from 1996 to 2021 it is found that there is no 	
notable trend in total suicides. More Suicides are committed in the year 2002 ( 9810) & 
Least Suicides are committed in the year 2015 ( 7692). 
 	
 	
 	
   	
Suicide Trends over the years (Male vs  Female)

92

93

94

95 	
 	
Prediction of Suicides 	
We  experimented  with  various  machine  learning  models  for  suicide  rate 	
estimation,  but  encountered  significant  challenges  with  low  accuracy,  impeding  our 
progress.  Nevertheless,  we  have  opted  for  the  Autoregressive  Integrated  Moving 
Average  (ARIMA)  model,  which  has  demonstrated  comparatively  good,  enabling  us 
to proceed with our prediction efforts.  	
 
 	
 	
 	
Cause	
Actual Suicides-	
2020	
Actual Suicides-	
2021	
Predicted Suicides-	
2022	
Predicted Suicides-	
2023	
Bankruptcy or indebtedness 180 183 176 172
Drug abuse/addiction 692 902 854 868
Failure in examination 43 53 56 58
Family Problem 3575 4549 4407 4535
Illness: 1933 2006 1996 1999
Love affairs 238 295 277 282
Marriage Issue 120 164 153 148
Others 1510 1082 1065 1060
Professional/ Career Problem 87 187 196 188
Unemployment 122 122 132 124
Total Suicides 8500 9543 9311 9436	
Prediction of Suicides using Time Series ARIMA Model

96

97

98 	
 
 
Major Findings of the Study 
 	
â¢ 	74% of Suicides in Kerala are committed by men. 	
â¢ 	Major cause of suicides in Kerala is family problem & illness. 	
â¢ 	Age group committing more suicide is 30-44 & 45-59. 	
â¢ 	Male  of  age  group  45-59  are  committing  more suicides,  in  case  of  female  it  is  15-
29. 	
â¢ 	Major  cause  of  suicide  of men  upto  age  59  is  due  to  family  Problem  but  above  60 
years is due to illness. 	
â¢ 	Major  cause  of  suicides  of  female  upto  44  years  is  family  problem  &  above  45  is 
illness. 	
â¢ 	There are no Specific trends in Suicides over the Years, but there is slight increase 
in suicide of men. 	
â¢ 	In 2015 there is a decline in suicide of all male age groups. 	
â¢ 	There is a decline in suicide of female age group 30-44 & increase in age 60 above. 	
â¢ 	Suicide due to bankruptcy & indebtedness is declining. 	
â¢ 	Most affected group of suicide by usage of drugs is male of age group 45-59 & 30-
44.  	
â¢ 	Suicide due to drug abuse is very low in case of female (Below 10 each year). 	
â¢ 	Male suicide due to drug abuse is increasing.

99 	
 
Conclusion 	
      Government  should  consider  comprehensive  mental  health  programs,  strengthening 
mental health infrastructure, and promoting awareness campaigns targeting different age 
and gender demographics. Proper interventions should address the root causes identified 
in  the  study,  including  family  problems,  illness,  and  the  rise  in  male  suicides  due  to 
drug  abuse.  Collaborative  efforts  between  government  agencies,  healthcare 
professionals,  and  community  organizations  are  essential  for  effective  suicide 
prevention  initiatives  in  the  state.  It  is  seen  that  suicide  is  done  due  to  one's  mental 
disorder or physical disability, as a result of excessive alcohol/drug use, due to excess of 
mental tension, due to love disappointment, or due to a disturbance in economic status. 
If there is a strong intervention of the government to adopt ways to follow the lifestyle, 
we  can  save  ourselves  from  the  big  challenge  of  suicide  that  is  facing  the  modern 
society.  While  this  project  was  in  progress,  an  article  was  published  in  the  editorial 
column  of  Mathrubhumi  daily  regarding  the  huge  increase  in  male  suicide  in  Kerala 
which reveals the relevance of this project. We still have a long way to go. We need to 
have  a  mentally  and  physically  healthy  generation.  May  this  report  increase  the 
attention  of  the  government  on  this  issue.  âCreating  Hope  Through  Actionâ  is  the 
triennial theme of the World Health Organisation for the World Suicide Prevention Day 
from 2021-2023. This theme serves as a powerful call to action and reminder that there 
is  an  alternative  to  suicide  and  that  through  our  actions  we  can  encourage  hope  and 
strengthen prevention.  
"When  you  feal  like  giving  up,  just  remember  the  reason  why  you  held  on  for  so 
long" 
âBecause if you kill yourself, youâre also going to kill the people who love youâ 
 
References 	
âSuicides in India - Basic Data Analysisâ - By Shavilya Rajput 
âPrediction of Suicide Causes in India using Machine Learningâ-   
By Imran Amin, Sobia Syed 
âAccidental Deaths & Suicides in Indiaâ â 
 Published by National Crime Records Bureau (NCRB)

100

101 	
 	
 	
 
 
 
 
 	
Analysis and Prediction of Monthly Percapita 	
Expenditure (MPCE) of a Family in Kerala 	
 	
 	
 
 
 
 
 	
Submitted by 	
Sri. Sreekumar G,  	
Research Officer

102 	
 
Introduction 
 	
The Household Consumer Expenditure Surveys of National Statistical Office (NSO) 	
are  the  primary  source  of  data  on  various  indicators  of  level  of  living,  pattern  of 
consumption  and  wellbeing  of  individuals  at  national  and  state  levels.  NSO  and  DES 
conducted NSS 68th round survey on Household Consumer Expenditure with 2:3 matching 
ratio sample size basis during July 2011 to June 2012. 	
The  NSS  consumer  expenditure  survey  aims  at  generating  estimates  of  household 	
Monthly Per Capita Consumer Expenditure (MPCE) and the distribution of households and 
persons over the MPCE range separately for States and Union Territories, and for different 
socio-economic  groups.  These  indicators  are  among  the  most  important  measures  of  the 
level  of  living  of  the  relevant  domains  of  the  population.  The  distribution  of  MPCE 
highlights the differences in level of living of the different segments of the population and 
is  an  effective  tool  to  study  the  prevalence  of  poverty  and  inequality.  These  numbers  thus 
enable the apex planning and decision-making  
 
Objectives 	
The MPCE play a critical role in the estimation of poverty line and living standard 	
of the people in the state. Moreover the vigilance department usually request to the DES to 
make available the average expenditure of some families involved in cases. In this context, 
this study has more importance. 
The main objectives of the study.  	
ï· 	To analyse the data of household consumer expenditure. 	
ï· 	To build a system capable of finding underlying patterns in the data. 	
ï· 	Predict the MPCE. 
 	
Literate review  	
A  similar  study  has  been  done  in  Forecasting  and  analysing  the  real  estate  market 	
can  evaluate  the  stability  of  the  real  estate  market.  On  the  one  hand,  it  can  facilitate  the 
government to carry out macro control on house prices and maintain the healthy and stable 
development  of  Chinaâs  national  economy.  On  the  other  hand,  the  prediction  of  house 
prices  can  also  provide  a  certain  basis  for  real  estate  investors  to  formulate  investment 
strategies and avoid losses.[4].  	
Another  paper  [5] 	  discusses  various work by  different  researchers  on  linear 	
regression  and  polynomial  regression  and compare their  performance  using  the  best 
approach to optimize prediction and precision.	 	
Through this project, I try to develop a machine learning model for predicting the 	
MPCE of a family in Kerala.  	
 	
Programming Language 	
I  use 	Python	  as  my  programming  language. 	It  is  a  versatile,  high-level  programming 	
language  known  for  its  readability  and  ease  of  use.  It  supports  multiple  programming 
paradigms,  including  procedural,  object-oriented  and  functional  programming.  Python's 
extensive standard  library  and  active  community  make  it  suitable for  diverse  applications, 
from web development and data analysis to artificial intelligence and automation.

103 	
 
Steps  	
â¢ 	Understand the Data 	
â¢ 	Handle Missing Values 	
â¢ 	Remove Duplicates 	
â¢ 	Correct Data Types 	
â¢ 	Standardize/Normalize Data 	
â¢ 	Handle Outliers 	
â¢ 	Address Inconsistencies 	
â¢ 	Check for Accuracy 	
 
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.	 	Pandas:	  A  data  manipulation  library  in  Python  that  is  often  used  to  clean, 	
transform, and manipulate data for machine learning. 	
2.	 	NumPy: 	A  Python  library  for  numerical  computing  that  is  often  used  for  linear 	
algebra and array operations in machine learning. 	
3.	 	Matplotlib:	 A Python plotting library that is often used to create visualizations of 	
data for machine learning. 	
4.	 	Seaborn:	 A data visualization library in Python that is built on top of Matplotlib 	
and provides a high-level interface for creating statistical graphics. 	
5.	 	Scikit-learn	:  A  popular  machine  learning  library  in  Python  for  tasks  such  as 	
classification, regression, and clustering. 	
6.	 	Statsmodels -	library for statistical modeling and analysis	  	
Data Set used 	
Pooled Data of 68	th Round Socio-Economic Survey conducted by National Sample 	
Survey (NSS) [July 2011 to June 2012] and dummy data. The model of data is given below. 
It has 6723 rows and 6 columns.

104 	
 
 
Table :1 
 	
 	
Missing values and duplicates are removed from the data set. 
Codes used in the data 	
Housed type: 	
Self-employed in agriculture - 1 
Self-employed in non-agriculture - 2 
Regular wage/salary earning - 3 
Casual labour in agriculture - 4 
Casual labour in non-agriculture - 5 
Others-9 	
Religion:	 	
Hinduism-1 
Islam-2 
Christianity-3 
Sikhism-4 
Jainism-5 
Budhism-6 
Others-9 	
Social group: 	
Scheduled tribes - 1 
Scheduled castes - 2 
Other backward classes - 3 
Others - 9	 	
The household type, religion and social group are all represented by codes, the system will 
be  treated  these  as  numerical  values.  So  they  are  expanded  along  each  code  with  two 
entries â â0 and 1â.

105 	
 
Table 2:  The Model of the Expanded Columns 	
 	
 
 
 
 
Figure 1 : Correlation Matrix of Expanded Table Data

106 	
 
 
Handling Outliers 
Figure 2 : Boxplot of Monthly Per Capita Expenditure 	
 	
 	
Outliers  are  data  points  that  deviate  significantly  from  the  rest  of  the  data  in  a 	
dataset. Whether or not to remove outliers depends  on the nature of the data and the  goals 
of the analysis. In some cases, it might be appropriate not to remove outliers. 
       Here  outliers  represent  valid  and  genuine  variability  in  the  data.    Removing  such 
outliers may result in the loss of important information. 
       Since  we  have  a  small  sample  size,  removing  outliers  may  significantly  impact  the 
representativeness of the data. 
      So we have to retain the outliers. 
 
Data at a Glance 
Figure: 3

107 	
 
Figure: 4 
 	
 	
 
 
 
 
Figure: 5 
 	
 	
The following machine learning algorithms are used.

108 	
 
Ordinary Least Square (OLS) 	
Ordinary Least Squares (OLS) regression is a statistical method used for estimating 	
the relationship between one or more independent variables and a dependent variable.  
OLS Regression Results             
========================================================	 	
Dep. Variable:          MPCE in Rs          R-squared:                    0.151 
Model:                      OLS            Adj. R-squared:            0.147 
Method:                    Least Squares          F-statistic:                     38.50 
Date:                         Mon, 18 Dec 2023    Prob (F-statistic):          1.70e-211 
Time:                        22:21:13             Log-Likelihood:            -61425. 
No. Observations:     6724       AIC:       1.229e+05 
Df Residuals:            6692      BIC:       1.231e+05 
Df Model:                  31                                          
Covariance Type:      non-robust                                          
======================================================== 
 
R-squared and Adjusted R-squared:	 	
ï· 	R-squared  (RÂ²):	 This value (0.151) represents the proportion of the variance in the 	
dependent variable (MPCE in Rs) that is explained by the independent variables in 
the  model.  Approximately  15.1%  of  the  variability  in  MPCE  can  be  explained  by 
the model. 	
ï· 	Adjusted R-squared:	 This is a modified version of R-squared that accounts for the 	
number of predictors in the model. The adjusted R-squared (0.147) is slightly lower 
than R-squared but still provides a measure of goodness of fit. 	
 
F-statistic:	 	
The  F-statistic  (38.50)  is  used  for  testing  the  overall  significance  of  the  regression 	
model.  A  high  F-statistic  suggests  that  at  least  one  independent  variable  is  significantly 
related to the dependent variable. The F-statistic is relatively high. 
Prob (F-statistic):	 	
The  p-value  associated  with  the  F-statistic  (1.70e-211)  is  extremely  low.  This 	
indicates  that  the  overall  model  is  statistically  significant,  meaning  that  there  is  strong 
evidence that at least one independent variable is related to the dependent variable. 	
In summary, the model  has statistical significance (low p-value for  F-statistic), but 	
the  R-squared  value  is  relatively  low  (15.1%),  suggesting  that  the  model  explains  only  a 
modest proportion of the variance in the dependent variable. Additionally, we may want to 
explore  potential  improvements  to  the  model	,  such  as  adding  relevant  variables  or 	
considering alternative model specifications. 
To get a better result I try the next model 	linear regression	 	
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).

109 	
 	
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. 	
We  have  more  than  one  independent  variables,  we  used  multiple  linear  regression 	
which is expressed as follows. 	
Y=B0+B1X1+B2X2+B3X3+B4X4+Ui 
Where Y=MPCE of the family 	
                X1=Household size 
                 X2=Household type 
                  X3=Religion 
                   X4=Social group 
    	            B0,B1,B2,B3,B4 are the  Regression Coefficients  	 	
                Ui=Disturbance term(Error term) 
Data Partitioning 	
The  Entire  dataset  is  partitioned  into  2  parts:  80%  dataset  is  used  for  training  the 	
model and remaining 20% data is used for test the model. 	
 
 	
 	
Results and analysis 	
 
Figure: 6 	
 	
R-squared: 0.12167642964911363 
Red:  Predicted  
Blue:  Actual

110 	
 
Figure 7 : 	The scatter plot between actual values and predicted values. 	
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Regression plot of the model. 	
A  regression  plot  is  useful  to  understand  the  linear  relationship  between  two 	
parameters. It creates a regression line in-between those parameters and then plots a scatter 
plot of those data points. 
Figure: 8

111 	
 	
Conclusion 	
 
Now with the help of above Linear Regression model, we can predict the MPCE of a 
family if the characteristics (household size, household type, religion and social group) 
of that family are given. 	
The  model  is  about  88%  accurate  to  predict  the  MPCE.  Statistical  model  of  the 	
project  provides  a  useful  tool  for  predicting  the  MPCE.  It  does  not  depend  on  any 
economic  parameter.  No  such  data  are  available  from  the  socio-economic  survey 
conducted  by  the  NSO.  So  further  set  of  data  linked  with  economic  conditions  of  the 
families  are  necessary  for  predicting  MPCE  which  reveals  the    real  picture  of  the 
expenditure  of  the  families.  The  potential  of  machine  learning  techniques  for  predicting 
MPCE under variety of scenarios requires further research and analysis. 
 
 
REFERENCE: 
 	
1.	 	Report on Household Consumer Expenditure in Kerala, Based on NSS 68	th Round 	
July 2011- June 2012 Central and State sample pooled data, Department of 
Economics & Statistics, Government of Kerala. 	
2.	 	Report on Household Consumption of various Goods and Services in Kerala, (NSS 
66	th Round July 2009- June 2010) Department of Economics & Statistics, 	
Government of Kerala. 	
3.	 	Various websites. 	
4.	 	Research on Prediction and Analysis of Real Estate Market Based on the Multiple 
Linear Regression Model, Sheng Bin , 2022 	
5.	 	A review on Linear regression comprehensive in machine learning, Dastan Maulod, 
Adnan N AbdulAzeez 2020

112

113 	
 	
 	
 
 
 
 
 
 	
Times Series Forecasting Study 	
for Death (1962 - 2020) using AI 	
 	
 	
 
 	
 
 	
Submitted by 	
Smt. Yamuna A.R,  	
Deputy Director

114 	
 	
1.	 	Introduction: 	
  Time series forecasting is a crucial aspect of data analysis, especially in the conte
xt of understanding and predicting trends in mortality rates over the years. In this study, we
 employ Artificial Intelligence (AI) techniques to analyze and forecast death data spanning 
from 1962 to 2020. The objective is to gain insights into historical patterns and trends and 
use AI models to predict future mortality rates. Predicting future deaths is a critical challen
ge in public health and demography. Understanding mortality trends helps governments, he
althcare providers, and policymakers to allocate resources effectively. 
 	
2.	 	Data Collection: 	
  The dataset used in this study encompasses death statistics from the year 1962 to 
2020. The data is sourced from civil registration database, and it includes various demogra
phic indicators related to live births, deaths, maternal deaths and infant deaths from the yea
r 1962 to 2020. 
 	
 	
3.	 	Preprocessing: 	
  To prepare the data for time series forecasting, preprocessing steps are implement
ed. This includes handling missing values, converting data into a time series format, and ad
dressing any anomalies or outliers detected during the EDA phase. For our time series fore
casting analysis, we focus exclusively on the  "Year" and "Death" columns within the datas
et. This targeted approach allows us to streamline the preprocessing steps and concentrate 
on the essential components relevant to forecasting. The "Year" column is converted into a

115 	
 
datetime format and set as the index, and any missing values in the "Death" column are add
ressed, ensuring a complete and coherent time series. 	 	
a)	 	Handling Missing Values:  	
  The dataset is inspected for any missing values in both the "Year" and "Death" co
lumns.Appropriate strategies, such as forward-filling, are employed to address and fill in m
issing values, ensuring a continuous and complete time series. 	
b)	 	Converting Data into Time Series Format: 	
   The "Year" column, originally presented as a standard numerical format, is conve
rted into a datetime type. The "Year" column is designated as the index of the dataset, trans
forming it into a time series format. This step is fundamental for chronological data analysi
s. 	
 	
 	
c)	 	Addressing Anomalies or Outliers Detected during EDA: 	
     Anomalies and outliers identified during the Exploratory Data Analysis (EDA) p
hase are specifically targeted for resolution.Techniques like Z-score analysis or other statist
ical methods are applied to detect and address outliers, ensuring that extreme values do not 
unduly influence the forecasting models. 
 	
d)	 	Visualizing Preprocessed Data: 	
 
  The preprocessed time series data is visualized to confirm that the conversion, ha
ndling of missing values, and outlier resolution have been successfully executed. Plots and 
visual representations help in assessing the effectiveness of the preprocessing steps and un
derstanding the data distribution.  Before delving into time series forecasting, an EDA is co
nducted to understand the distribution, trends, and seasonality in the death data. This phase

116 	
 
 involves visualizing the time series, identifying any outliers, and exploring any discernible
 patterns.  	
 
 	
  We employ outlier detection techniques, such as the boxplot method, to identify a
nd handle any extreme values that might impact the forecasting models. The resulting prep
rocessed time series data, visualized forms the foundation for our subsequent time series fo
recasting analysis.. 
  By undertaking these preprocessing steps, the dataset is optimized for time series 
forecasting, ensuring that it aligns with the prerequisites of accurate and reliable predictive 
modeling. 	
4.	 	Model Selection: 	
  Several AI models are considered for time series forecasting, with a focus on thos
e designed for sequential data. Common choices include Autoregressive Integrated Moving
 Average (ARIMA), Seasonal-Trend decomposition using LOESS (STL), and more recentl
y, machine learning techniques like Long Short-Term Memory (LSTM) networks, which a
re well-suited for capturing long-term dependencies in sequential data. 	
a)	 	Train-Test Split: 	
  The dataset is split into training and validation sets. Typically, earlier years are us
ed for training, and later years are reserved for validation.

117 	
 	
b)	 	Scaling Data: 	
  Feature scaling is applied to normalize the data. Min-Max scaling or Standard sca
ling is  used for LSTM model. 	
c)	 	Model Training: 
1.	 	Long Short-Term Memory (LSTM) networks 	
  Initially, the Long Short-Term Memory (LSTM) network is selected as the model
 for time series forecasting. The chosen AI model undergoes training using historical death 
data. 	
Validation and Evaluation:	 	
  The  trained  model  is  validated  using  the  validation  set  to  assess  its 
accuracy.Various  evaluation  metrics  such  as  Mean  Squared  Error  (MSE)  or  Root  Mean 
Squared  Error  (RMSE)  are  computed  to  gauge  model  performance.  Mean  Squared  Error: 
269443718.62611395 and RMSE (Root Mean Squared Error):16414.7408

118 	
 	
 	
   
 	
2.	 	Autoregressive Integrated Moving Average (ARIMA) 	
  Next,  we  explore  the  ARIMA(3,2,5)  by  inspecting  ACF  and  PACF  plots  after 
differencing  the  time  series  data  inorder  to  make  it  stationary.  The  ARIMA  model  stands 
for  Auto  Regressive  Integrated  Moving  Average,  and  the  numbers  (p,  d,  q)  represent  the 
order  of  the  autoregressive,  integrated,  and  moving  average  components,  respectively.The 
ARIMA(3,2,5)  model  represents  an  AutoRegressive  Integrated  Moving  Average  model 
with three autoregressive (AR) terms, two differencing (I) terms, and five moving average 
(MA) terms. 	
Validation and Evaluation:	 	
  The AI model's forecasting performance is analyzed, and predictions for future de
ath counts are generated. Visualizations of the forecasted time series data are created to de
monstrate the model's ability to capture trends and patterns. 
  RMSE (Root Mean Squared Error): The RMSE value is 8537.676. This value rep
resents the average magnitude of the error. MAE (Mean Absolute Error): The MAE value i
s 6378.316.MPE (Mean Percentage Error): The MPE value is 1.049009. A MPE close to ze
ro is generally desirable, indicating a low average percentage difference between predicted 
and actual values. MAPE (Mean Absolute Percentage Error): The MAPE value is 4.45826
5. Like MPE, a lower MAPE is preferred. This metric provides the average percentage diff
erence between predicted and actual values. MASE (Mean Absolute Scaled Error): The M
ASE value is 0.8674745. A MASE close to 1 suggests that the ARIMA model is performin
g well compared to a naive forecast.

119 	
 
 	
Year 	Actual 	Fitted 	
1962	 	110456	 	110406.6026	 	
1963	 	106667	 	106823.664	 	
1964	 	106496	 	104391.7208	 	
1965	 	97709	 	102517.5753	 	
1966	 	102245	 	96516.79461	 	
1967	 	94552	 	97442.66006	 	
1968	 	92669	 	93397.43246	 	
1969	 	87186	 	88089.43266	 	
1970	 	69745	 	85712.07919	 	
1971	 	91272	 	65851.55688	 	
1972	 	102515	 	86558.98614	 	
1973	 	105721	 	98135.11035	 	
1974	 	113447	 	105856.0146	 	
1975	 	131473	 	120527.1884	 	
1976	 	133602	 	127656.2841	 	
1977	 	119113	 	137034.0829	 	
1978	 	114434	 	122557.2126	 	
1979	 	117961	 	117604.5951	 	
1980	 	118140	 	114312.5699	 	
1981	 	122904	 	116046.4438	 	
1982	 	119827	 	124393.2793	 	
1983	 	119414	 	121717.9708	 	
1984	 	120841	 	120398.1615	 	
1985	 	134230	 	121339.0565	 	
1986	 	133654	 	131407.0327	 	
1987	 	141047	 	136527.893	 	
1988	 	146596	 	141929.737	 	
1989	 	146194	 	149869.0503	 	
1990	 	147551	 	145786.5311	 	
1991	 	156919	 	151310.5592	 	
1992	 	162644	 	155953.6241	 	
1993	 	161403	 	163929.2134	 	
1994	 	163711	 	164500.3616	 	
1995	 	162868	 	166462.3883	 	
1996	 	172103	 	164580.8401	 	
1997	 	166428	 	171345.9153	 	
1998	 	185788	 	170266.774	 	
1999	 	186828	 	184320.4002	 	
2000	 	178795	 	191355.8657

120 	
 	
2001	 	182059	 	180505.222	 	
2002	 	184597	 	188203.0949	 	
2003	 	194264	 	181372.2592	 	
2004	 	199017	 	195513.4959	 	
2005	 	204157	 	201646.9465	 	
2006	 	219094	 	206550.8613	 	
2007	 	238691	 	221986.8416	 	
2008	 	221769	 	237337.7915	 	
2009	 	238691	 	230444.5631	 	
2010	 	238864	 	240417.4254	 	
2011	 	245002	 	242543.0694	 	
2012	 	239982	 	242352.7981	 	
2013	 	260195	 	250253.2754	 	
2014	 	248242	 	255357.5514	 	
2015	 	252576	 	257366.9326	 	
2016	 	256130	 	251679.5837	 	
2017	 	263342	 	263036.3889	 	
2018	 	258530	 	259219.4433	 	
2019	 	270567	 	266949.5301	 	
2020	 	250983	 	269699.9342	 	
 	
 	
 
5. Conclusion:  
  This time series forecasting study for death data using AI provides valuable insig
hts into historical trends and future predictions. After comparing the Root Mean Squared E
rror (RMSE) of the two models, namely the Long Short-Term Memory (LSTM) and the A
RIMA(3,2,5), the ARIMA(3,2,5) model demonstrates superior performance and is selected
 as the preferred model. The decision is based on the evaluation metric, where a lower RM	
Forecast  of Death upto three years	
year	
1960 1970 1980 1990 2000 2010 2020	
100000 200000 300000	
Actual
Fitted

121 	
 
SE signifies better accuracy. In this case, the ARIMA(3,2,5) model exhibits a lower RMSE,
 indicating that it provides more accurate predictions compared to the LSTM model. Theref
ore, the ARIMA(3,2,5) model is deemed the better choice for the given time series forecast
ing task.The chosen AI model demonstrates its efficacy in capturing patterns within the dat
a, offering a tool for policymakers and healthcare professionals to anticipate and address ev
olving mortality trends. Further research and continuous model refinement can enhance the
 accuracy and applicability of such forecasting methodologies. 	
Forecast of Death upto three years obtained from ARIMA(3,2,5) 	
 	
Year	 	Forecast	 	
2021	 	258927 	
2022	 	257882 	
2023	 	262216 	
 
5.	 Libraries and Tools Used 	
 	
6.	 Scikit-learn: Used for machine learning model implementation. 	
7.	 Pandas: Utilized for data manipulation and analysis. 	
8.	 Matplotlib and Seaborn: Employed for data visualization. 	
9.	 TensorFlow: Applied for deep learning tasks. 	
10.	 Jupyter  Notebooks:  Used  as  the  primary  environment  for  code  development  and 
experimentation

G
o
v
e
r
n
e
m
n t   o f   K e r a l a
1
9
5 1
R
E
P
O
R
T
 
O
N
 
P R E L I M I N A R Y
C
R
O
P
 
C
U
T
T
I
N
G
 
S
U R V E Y   O N   P A D D Y
I
N
 
P
U
N
J
A
 
F
I
E
L
D
S   O F   K U T T A N A D
A
N
D
 
N
E
I
G
H
B
O
U R I N G   A R E A S        
G
o
v
e
r
n
e
m
n
t
 
o
f
 
T
r a v a n c o r e -   C o c h i n
B
O
A
R
D
 
O
F
 
S T A T I S T I C S
C
e
n
t
r
a
l
 
R
e
s
e
a r c h   I n s t i t u t e
T
r
a
v
a
n
c
o
r
e   U n i v e r s i t y
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

Capacity Building

DATA ANALYTICS

OPEN DATA

DASHBOARDS

OTHER OPEN DATA PORTALS

Publications

Category: General Publications

Transcript
