International Journal of Recent Technology and Engineering (IJRTE)

ISSN: 2277-3878, Volume-8 Issue-5, January 2020

Socio-Economical Status of India using Machine

Learning Algorithms
V. Balasankar, P. Suresh Varma

 poverty and the state wide poverty were presented.

Abstract: Data is everywhere and lots of data is openly Algorithms like cluster analysis, correlation, decision
available to people. We can analyze this data to find the hidden trees, K-means, are studied on census and other related
and unnoticed information to use it purposefully. One important economic data. In this paper we have implemented K-means
source of information is census data and it provides data related to
and kmediods clustering algorithm on census data to find the
the people living in a country. Analyzing such data is useful for
knowing the socio economic status of the country. Data mining economic status of people in India. Socio economic
and machine learning techniques can be used to analyze such indicators like poverty, gender distribution in work, literacy
large volumes of data. In this work Indian census 2011 is analyzed rate for male and female, percentage of people working in
and identified the socio economic status of different states of different sectors, are measured.
India. To identify the social status of each state we studied literacy Section II presents literature review and section III
rate, categories of workers in different fields, gender wise working
presents the detailed methodology for classification section
population. To identify economical status like people living below
poverty and above poverty we used clustering techniques of IV presents results final section presents the conclusion.
machine learning. At first we pre-processed the data and later
correlation based feature selection was applied, and on that result II. LITERATURE SURVEY
k-means and k-mediods clustering methods were implemented
independently. Finally the clusters are evaluated to see the Author (s) can send paper in the given email address of the
performance using confusion matrix. The final results show that journal. There are two email address. It is compulsory to send
k-mediod has better performance than K-means. paper in both email address. Ronit nirel et.al[2009] described
about sample surveys and censuses. The author described the
Keywords: Machine Leaning, k-means, K-monodies, new trends in census preparation methodologies. Population
Socio-Economical, poverty census is a very important official statistical data, on which
many analyses are built. it should be defined properly and
I. INTRODUCTION
accurate data about individuals place of residence and other
demographic characters need to be collected. It is a complex
Governments often want to know the socio economic task and a costly procedure. There are many types of errors
conditions of each region, so it collects data from various and one of the important errors was coverage error which is
sources and analyzes this data. Inequalities in income are one further divided in to under coverage and over coverage. This
of the main problem that government want to solve. By study tells the eligibility status of an individual. Due to
reducing income inequalities we can achieve balanced social advancements in technology there is big influence on census
development; It also increases national income and provides preparation with GPS, Mobile phones and advancements in
political stability in the country. For this purpose internet. Regarding data collection there are two methods are
India-districts-census data was used to study the socio there one is record linkage procedure which uses information
economic developments in the specific regions. from administrative sources and the other one is sample
India-districts-census contains large volumes of data with surveys . Estimation of coverage errors by using sample
many attributes. But many of these attributes are not filled survey, use of sample data to evaluate statistical adjustment
and may not contain a value. Such sparse data is a common in of census counts were described. A different type of census
many applications. The dimensionality of objects may be that is based on continuous sample survey called rolling
very high like in census data. Such sparse data need to be survey implemented by France was also described. Finally
handled properly in order to improve the query processing sample surveys in conjunction with census and usage of long
and minimizing the storage, it is important to find active form and short forms as a data collection was described.
attributes and their correlations with other objects there by C.Hakim [1984] reviewed the new trends in census data
predicting regional economy .So many data mining and dissemination in particular data dissemination in Great
machine learning techniques have been implemented to clean Britain. Author discussed about many key issues like
, Analyze data and to predict the socio economic status of importance of census analysis by different people and the
people in a region and cluster have been created based on the users of these analysis and the need for summaries and
commentaries of census tables, guides and indexes for fast
Revised Manuscript Received on January 20, 2020 searching of the content to help the different users to
* Correspondence Author understand the census statistical tables, consultations with
V.Balasankar,*Research Scholar, Department of CSE, Adikavi
Nannaya University, A.P., India,Associate Professor, Department of
users for census data design, data dissemination centres like
CSE,Aditya college of engineering, Surampalem, Andhra Pradesh, India. census regional office, central
E-MailL balasankar.v@gmail.com government departments,
independent research institutes,
Dr. P. Suresh Varma, Professor, Department of CSE,Adikavi
Nannaya University,Andhra Pradesh, India universities, commercial

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3804 & Sciences Publication
 Socio-Economical Status of India using Machine Learning Algorithms

agencies and Also discussed about census by products and very close to the global means, high income areas of city
the need for census based area classification, publishing were generated and the zones were analysed This work was
sample data for public use and the related confidentiality mainly provides the analysis about the relationship between
issues, alternatives to the census like national surveys were the population and transportation trips in MZMV area.
discussed. Even though there are developments in census Jose cazal et al[2015] Explored predictive models for
data dissemination. And alternative sources of information, economic system. It was observed that selection of what data
the basic published volumes are sufficient for general to be included and what to be discarded is a big challenge
analysis. .There are some variables that are complex and many factors
Bin sheng et.al [2010] presented the importance of data affect these variables. Data mining techniques scientifically
mining technique to analyze census data. Census data proven and gives reliable results in analyzing, predicting
contains rich set of information and many meaningful socio economic studies. In this work data mining techniques
insights are hidden in it. CART (Classification and regression were proposed as a valid option which is better than
trees) is a decision tree based classification model, using this traditional econometrics methodology. SEMMA model was
model census data was analysed and predictions were made used in data mining which stands for Simple, Explore,
on socio economic conditions of the people. Decisions trees Modify, Model and Assess. Here based on type of data set
of classification gives high accurate prediction results, and (time series or cross sectional) predictive algorithms were
the advantage is the results are easy to understand. CART has chosen. Experiments were done on two cultures namely data
sound theoretical foundation , it has three phases like tree modelling culture (here data is generated by stochastic given
building, tree pruning, and tree estimation. Using c++ model) and algorithmic model culture (which assumes the
language CART model was implemented on fifth census data data as complex and unknown). Author did analysis and
of Chengy and Laixi. GINI index was used as a evolution prediction using both traditional econometrics with E-views
function in this work. Even though CART is a recursive tool and data mining techniques with EMMA tool. The
algorithm, non-recursive version was used to improve the results show that data mining techniques were more effective
performance. The main aim of this work was classifying the and efficient than the techniques of econometrics.
residents in Chengy and Laixi into four categories like poor, Joon heo et al[2014] proposed user driven economic data
general, better and best. The results showed that middle level analysis by using a mobile app. The user sends the data and
per capita income people were more in this area. Based on the the analysis parameters through mobile phone to the server.
results local authority can plan for economic development. The server uses big data and mathematical algorithms to
Jian ming et.al [2018] presented the implementation of perform analysis on given parameters. Normally Big data
artificial neural networks on economical and technical data of analysis is done by only few companies because they can
mining enterprise. Generally mining enterprise data is afford for it and user-oriented analysis were generally
multi-dimensional and nonlinear. One of the important neglected. This framework contains two entities one is server
indicators of mining enterprise is mineral products sales price and the other one is user application running in the mobile
data. Due to some technical limitations in the environment phone. Stock data of 8 countries over a period of 33 years
the geological data was lost and the author reconstructed this were used for this work. This data was first cleaned and next
data using artificial neural networks and geo statistics. Here it was processed. Android SDK was used to build this mobile
back propagation algorithm of artificial neural network app. At the server side three algorithms were implemented
model was used to predict the mineral product price. Neural (minimum spanning tree, principal component analysis and
network was created with a single hidden layer and three clustering). User selects economic parameters from the
–layered neural network with 5input neurons and 1 output mobile and transmits it to the server and the server sends the
neuron was built. For reconstruction of geographic data results back to the mobile device.
artificial neural networks and geo statistics were used. The Sharath R et al[2016] studied and analysed
predictions showed that the model is strong and prediction socio-economical conditions from US household data. In this
accuracy is high, for geological missing data the prediction work the size of dataset is huge i.e., 3.5 million household
results and interpolation were reliable. information. The major conclusion was income of
Maria Beatriz Bernabe Lacranca et.al.[2004] described a individuals decides various aspects of life like education,
statistical classification approach on subset of data from the health, standard of living, household decisions, and economic
XII house hold and population census data of metropolitan status and so on. In this work five modules were implemented
zone of the maxico valley(MZMV) to present the properties to study the importance of income domain. The five modules
of the population data. The main idea was classification of are gender distribution in occupation (male vs female),
zones in MZMV region. in this work K-Means procedure was education-salary relation (higher degrees and their income vs
implemented to create the clusters. To analyse this population professional degrees and their income), economic hierarchies
data cluster analysis was applied. To represent the area of to find the economic classes using classifiers, Benford’s law
study 57 variables were used which came from correlation of US income (outlines the frequency distribution in many
analysis using k means the number of clusters maintained datasets), mean and median of income distribution across all
going from 8 to 70 range. K means needs the n value for the the states. These economic hierarchy predictions are useful in
number of clusters to be formed in advance and it also needs many areas like planning houses for poor and middle class
k value for k initial centroid in advance. The statistical R people, better pension plans for retired people, planning for
software was used to classify 4925 census zones. Zones various welfare programs to
belong to clusters having greater size, lowest average in all poor and so on. To conduct this
variables, zones with lowest population ,zones mean values work five tools are used, they

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3805 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-5, January 2020

are Hadoop, Java1.7, Python, R, and Pig. The data was first was implemented. Final results showed that manufacturing,
normalised by eliminating less important attributes, null building trade, wholesale and retail markets are more
values. After normalisation economic hierarchies were developed than other trades.
created using K-means and predictions of those economic Avery sandborn et.al.[2016] described the relationship
classes are done using classifiers. Three classifiers were between special features derived from high resolution
applied to improve the efficiency .All of them performed satellite imagery and Census data of Accra Ghana. Special
nearly same on the dataset. The accuracy by those classifiers features are the metrics that analyze pixel groups for
were as follows: describing geometry, orientation, patterns of objects in an
image. Such special features can be used to find housing
conditions and living standards in a city. To see the
Table 1: Classifiers and Accuracy
Classifiers Accuracy
association between demographical variables and special
features five special features (panTex,Line support regions,
Naïve Bayes 48.13%
histograms of oriented gradients, Fourier transform ,local
C4.5 51.3%
binary patters) were selected and extracted from image and
Boosted C5.0 53.7% then related to census variables. This method was proposed
as a alternative methodology for census data. The special
Finally using relevant attributes demographic graphs feature and spectral information normalized difference
were plotted. Octavio juraz Espinosa et al[1999] described vegetation index.(NDVI).were computed and correlated to
visual techniques for input/output data. These visualisations census data and there exists a high correlation between LBP
are necessary because the data was represented in matrix and census derived variables. final results shows that the
form. So data navigation becomes a problem because of special features can be used to find the socio economic
screen limitation. Here the techniques were designed based conditions of the population.
on user tasks like querying about interaction of two sectors, Ferdin joe j et.al[2011]. Differentiated Horizontal,
labelling data points, matrix area magnification, comparing Vertical representation of data set with a new representation
two industries based on the goods they produced, comparing called Hover(Horizontal over Vertical). In many applications
two sectors based on the co efficient values, finding patterns sparse data is common issue. This sparse data is not managed
for matrix, modifying values and re-computing the matrix. In well in both Horizontal and Vertical models in the context of
this work better visualisation techniques were created to performance, storage, and query processing .The author
represent the economic input output data. The main need of worked with a new method called Hover on Census data and
analysis is to study the interdependencies between industries ecommerce data in sparse form. Hover representation
in regional economy. The data was maintained in four contains two steps generating correlation table and
different matrices like make matrix(commodities produced generating subspaces the sub spaces are generated from the
by industries), total requirements matrix(direct and indirect correlation table in a Heuristic manner. In this work the
interactions between industrial sectors),use matrix(inter author used rapid miner tool and measured subspaces, space
industrial activities and commodity inputs for industrial usage, execution time, running time in a systematic way and
production),direct matrix(based on use matrix and total the and the changes in parameters with the schema changes
output). All these matrices are displayed with the help of a are analysed and differences are observed.
window using a pixel for each cell , colours were assigned Zhuang et.al.[2014] dealt with the analysis of regional
based on its category. Each window is partitioned into three. economic indicators. Traditional economic methods are not
First one is a large window which contains matrix, second the effective in finding the factors influencing economy. In this
bottom part contains the detailed information and third left work the analysis on key factors of regional economy were
window represents the data to be visualised. Apart from these done with the help of k-means and CADD algorithms. High
matrices economic IO data was also represented as –dimensional data is sparse in nature, when processing such
geographical information. This way the visualisation makes data based on distance and density the clustering was not
it easy to perform analysis on economic input output data. efficient. In order to improve efficiency of clustering
Yin Cai et.al.[2010] studied online analytical mining for weighted CADD was proposed by the author. This partition
analyzing regional economic data. Regional economic data is the cluster based on adaptive density reachable ideas. Here
gathered from statistical data. As per this work old statistical the natural, economic attributes of the regional economic data
data was maintained in the form of word or excel format such was clustered using k-means but the results were not ideal, so
a data is structured in non hierarchical manner. This creates a CADD algorithm was used which reduced the total
problem in analysis and research. The regional economic data dimensions. Finally comparative analysis was done on
contains large number of industries in various regions. In this Chinese regional economics.
work the data was analysed with the help of online analytical
mining (OLAM) it was designed with the help of Data III. METHODOLOGY
warehouse, Analytical processing and data mining. The are
In this work we studied the socio economic status of
three main components namely data layer, middle layer,
various states and union territories of India.
client layer. data layer was built with MS SQLServer 2005 as
We have taken India-district-census 2011 data set for this
data warehouse. Middle layer built with MS Analysis service
work. This data set contains 640 rows and 118 attributes of
and created data cubes, metadata, OLAP engine, data mining
different districts from 28 states and 7 union territories.
engine. At the client layer web application was developed
Except three columns all the columns contain numerical
using OLAP service and different operations like data cube
values.
browsing, rotation, slicing, and drilling can be done. Here
The data in this data set
zhenjiong province economic data was used and
contains values in aggregated
multidimensional cubes are created and clustering algorithm

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3806 & Sciences Publication
 Socio-Economical Status of India using Machine Learning Algorithms

form and the usage of this data is limited. Form this data we
want to analyze socio economical condition of India. The
proposed work is divided into 2 main modules

1) Analysis of social conditions

2) Economical conditions like measuring people living
in bellow and above poverty.

Analysis of social conditions includes the following sub

modules.
1) Finding different categories of workers state In this work python has been used to study socio
wise. economical conditions. All the required modules are
2) Finding gender wise working group for all states. imported for filling missing values, normalizing data, for
3) Finding gender wise literacy rate for all states. statistical calculations and plotting data.
In this module fist data set was verified for missing
3.1 social economical condition values. And the missing values are filled with mean value of
that attribute.
Different categories of workers
The data was scaled using min max scaling which
From this data set various working categories are found. transforms features in the given range here we have scaled in
Like percentage of people working in Cultivation, the range of 0 to 9.
Agriculture, Household and Others from the total working Creating clusters using K-means and K-mediods
population state wise. And all these categories are plotted on
graph state wise This experiment was divided into two steps:
Gender wise working group 1) Finding most correlated or significant variables
2) Creating clusters using k means and kmediods
Here from the data set we have extracted total males and
females state wise, and total male workers and female The first step is common for both k means and kmediods
workers state wise, and found the percentage of males and that is finding most correlated or significant variables. Here
females in different working area. This tells us the we have used Pearson-correlation-coefficient method on this
employment conditions for males and females. It was found data set. This has given the most correlated 12 features out of
that the percentage of male and female workers in 118 attributes. There is 0ne important feature called
LAKSHADWEEP it is 46.24,11.4 respectively, and this is ‘Households_with_TV_Computer_Laptop_Telephone_mobi
the minimum value in this working group when compared le_phone_and_Scooter_Car’ which talks about the living
with all other states. The average working group of India is standards of people.
54.37 and 31.08 for male and females respectively. For these 12 features we have calculated mean values
Gender wise literacy rate for all 640 districts of 28 states and 7union territories. From
these mean values we find class label. This class label
Here we analyzed state wise male and female literacy identifies poverty status. As the dataset does not provide class
rates. label for each district the following method was used.
Here we found male and female literacy rate by
calculating percentage of male and female literates from the 1) first we find the difference between maximum and
total population of male and total population females state minimum mean values of all districts.
wise respectively. This gives us male and female literacy rate 2) 20% of this difference is added to the minimum mean
for each state. It was found that the percentage of male and value and placed in variable called pvalue.
female workers in BIHAR it is 58.23,41.94 respectively, and 3) finally we compared mean value each district with the
this is the minimum value in this working group when pvalue . If the district’s mean value is greater than pvalue
compared with all other states. The average literacy rate of then the class label is 1 other wise it is 0.
India is 73.31,61.68 for male and females respectively.
After this K means and kmediods clustering techniques
3.2 Economical conditions Of the people are applied independently to find the state population bellow
and above poverty line. And these clusters are compared with
The objective this module is to find the poverty in the class label that was calculated in the previous step using
India,and finding population above poverty line and bellow confusion matrix and the results are presented. The entire
poverty line. Here on the dataset we applied K- means, procedure is detailed here under. Pearson-correlation
kmediods clustering algorithms. The dataset is a open dataset coefficient method :
and it contains partial data about 28 states and 7 union This method is a measure of strength which tells us how
territories. With the available data and with some two variables are linearly associated with each other. In our
assumptions clusters were created for states above and below data set we have total 118 columns and 640 rows. We have
poverty. one attributecalled ‘Households with TV Computer Laptop
The method for creating clusters is represented with the Telephone mobile phone and Scooter Car’ which is an
help of flow chart as given below. important attribute that we have used to find most important
features that can influence the
poverty condition.

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3807 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-5, January 2020

Pearson-correlation-coefficient is defined as A mediod is nothing but any object or data point of the
cluster, whose average dissimilarity to all other bojects or
data points in the cluster is minimal.

Here the basic idea is first compute the k representative

objects and call them as mediods. After set of mediods are
found each representative object of the data set is assigned to
one of the nearest mediod. Which means object i is put into
cluster ci when the mediod mci is nearest than any other
With this we found attributes. Based on these 12 highly method mj
correlated features we find the clusters using K-means.
k means clustering Implementation: The algorithm proceeds in two steps
1) Build: This is a initialization step, select k random data
The first two rows of the dataset are chosen as the initial points out of n points as the initial medicos.
centroids and found the euclidean distance from the first row 2)Swap: if the objective function can be minimized by
to all the other rows and stored in a variable Temp1 and found swapping mediod with any other data point, than swapping is
performed. This is repeated until the objective function can
euclidean distance from second row to all other rows and
no longer be minimized.
stored in a variable temp2. And tested whether temp1 is
K mediods algorithm is implemented as follows
minimum or temp2 is minimum .if temp1 is less than temp2
then our prediction is true and its value is represented as 1 Input:
and we are comparing this with actual result which is in class K:Represents the number of clusters
label. if it is false its value is 0 then we are incrementing false D: A data set consists of n objects
negative otherwise we increment false positive than current Output:
c1 value is put into x and y is the current row value. Now we A set of K clusters
are taking mean of x and y that become the new centroid . Method:
After this c1 is updated with nc1 which is a new centroid 1. The algorithm starts with random selection of the K
applied to test the remaining rows. Else if temp1 greater than representative objects as medoid points out of n data points.
temp2 we increment true positive by 1. If the actual value of 2. Associate each data point to the nearest mediod by using
classlabel is 1, else we increment true negative. We move c2 any common distance metric methods.
and current row in x and y variables respectively and taking 3. Randomly select any non-medoid object Orandom.
mean value of x y we will get the new centroid. We repeat all 4. Compute total cost S of swapping current medoid with
these steps until previous centroid and the newly found object Orandom .
centroid are equal. The objective function is defined as 5. If S>0, swap current medoid with the Orandom.
6. Repeat these steps until there is no change in the medoid.
2
K means=

K means [14] algorithm is defined as follows : IV. EXPERIMENTAL RESULTS

1. Take data points and cluster them in to k groups The social status of the states in India are represented as
where k is already defined. different categories of workers, gender wise working group
and gender wise literacy rate. The results are as follows. The
2. From the dataset choose k points at random as
percentage of male workers
cluster centers.
Are presented in 1(a), The percentage of female workers
3. Assign each data point to their closest cluster center Are presented in 1(b),1(c) represents country wise male
according to the Euclidean distance method. versus female literacy rate.
4. Calculate the mean value of all data points in each
cluster now this become the new centroid for the
cluster.
5. Repeat steps 2, 3 and 4 until the same data points are
assigned to each cluster in consecutive rounds.

Finally we are calculating the actual mean values of each

state and showing the same on pie chart.
K MEDIODS: K-mediod is also called as partitioning
around mediods method(PAM). It is a partitional algorithm
for clustering and it is a extenction to K-means algorithm. It
is more robust noise and outliers than K-means.
In k-means method mean values are choosen as centriods
for the cluster,whereas in K-medios the representative data
points are choosen as the mediods. K-Mediod is
interapretable as one of the data points is selected as mediod.

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3808 & Sciences Publication
 Socio-Economical Status of India using Machine Learning Algorithms

1(a) Percentage of male workers state wise

1(b) Percentage of female workers state wise

missing values are filled with the mean values of the
respective feature. The following diagrams represent this
issue. Fig. 2 (a) Represents missing values in the original
dataset. Fig 2(b)Represents data after filling missing values.

1(c) Male versus female literacy rate country wide

2) Finding different categories of workers state wise.

3) Finding gender wise working group for all states.
Fig.2.(a) represents missing values for some columns
4) Finding gender wise literacy rate for all states.

For experimental purpose Indian districts 2011 census data is

taken. The data set contains different states and districts
information. It contains total 640 rows and 118 columns. The
data set was first analyzed for missing values, after that the

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3809 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-5, January 2020

K-Means clustering is applied on the data set and the

data is made into two clusters namely bellow poverty and
above poverty. The following Figures represents state wise
poverty percentage.fig.3(a) shows the actual poverty in all
over India and fig.3(b) shows the experimental results for
poverty in India with k-means.

Fig.2.(b) represents data after filling missing values

fig.3(a) The actual poverty in India
After this Pearson correlation is found with respect to an
important feature called ‘Households with TV Computer
Laptop Telephone mobile phone and Scooter_Car’ which
describes the economic status of Households . After this the
following 12 fields are more correlated to this feature. Only
these features are used for cluster creation.

1)Other_Workers, 2)LPG_or_PNG_Households,
3)Households_with_Internet, 4)Households_with_Computer
5)Urban_Households, 6)Graduate_Education,
7)Households_with_Car_Jeep_Van
8)Households_with_Radio_Transistor9)Households_with_S
cooter_Motorcycle_Moped
10)Households_with_TV_Computer_Laptop_Telephone_m
obile_phone_and_Scooter_Car, Fig. 3(b) The experimental results for poverty in India
11)Households_with_Telephone_Mobile_Phone_Both with K-means
12)Ownership_Rented_Households' fig.4(a) shows the actual poverty state wise in bar
The above 12 fields are more correlated and they are diagram fig.4(b) shows the experimental values of different
important for calculating poverty. states in bar diagram.

Fig. 4(a) Actual Poverty State Wise

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3810 & Sciences Publication
 Socio-Economical Status of India using Machine Learning Algorithms

Fig. 4(b) experimental values of different states in bar diagram

When Kmediods algorithm applied on the 12 highly

correlated features the data is made into two clusters namely
bellow poverty and above poverty. The following Figures
represents state wise poverty percentage.fig.5(a) shows the
actual poverty in all over India and fig.5(b) shows the
experimental results for poverty in India with K-modiods.

Fig.5.(b) The experimental results for poverty in India

with Kmeans

fig.6(a) shows the actual poverty state wise in bar diagram

fig.6(b) shows the experimental values of different states in
bar diagram.

Fig.5.(a) The actual poverty in India

Fig.6.(a) Actual poverty state wise

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3811 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8 Issue-5, January 2020

Fig.6. (b) Experimental values of different states After K-mediods

The following table shows the compassion of k-mediods

and k means with respect to performance measures like Error ACKNOWLEDGMENT
rate, Precision, Recall, F-measure, Sensitivity, Specificity,
and Accuracy. We would like to thanks the Director and management of
Aditya college of engineering, College, Surampalem for give
the assistance and support for this work. We are very much
Table 2: K-Means and K-mediods Analysis
Measures of K-means (in K-mediod (in thankful to R&D department of Adikavi Nannaya University,
performances %) %) Andhra Pradesh, India for accepting and give their
Error rate 20 4
cooperation for this work. We are also thanking to Dr.
Precision 0 100 T.PanduRanga Vital for valuable suggestions towards this
Recall 84 100 research work.
F-measure 0 100
Sensitivity 0 15 REFERENCES
Specificity 84 100
Accuracy 81 97 1. Ronit Nirel and Hagit Glickman “Sample Surveys and Censuses”,2009
Elsevier, Vol. 29A,issue PA,p.g 539-565.
2. ⁠C .Hakim, “Data dissemination for the population census”,Social
As comparative analysis, the Kmediod performance is Science Information Studies (1984), vol 4,issue 4,p.g 273-282.
better than Kmeans for this dataset. The table 2 shows the 3. Bin Sheng,Sun Gengxin,“Data Mining in census data with CART”,
ICACTE 2010 - 2010 3rd International Conference on Advanced
detailed analysis about k-means and k-mediods that the Computer Theory and Engineering, Proceedings (2010),vol 3,p.g
Kmediod is giving better accuracy, specificity, recall, 260-264.
precision-measure and less error rate when compared to 4. Jian Ming, Lingling Zhang, Jinhai Sun,”Analysis models of technical
and economic data of mining enterprises based on big data
Kmeans. But for the sensitivity parameter, K-means is giving analysis”,2018 3rd IEEE International Conference on Cloud Computing
0 whereas Kmediod is giving 15. and Big Data Analysis, ICCCBDA 2018,p.g 224-227.
5. Maria Beatriz Bemabe Loranc,Ramiro Lsez,”Application of
Nonsupervised Classification to Population Data”,2004 1st International
V. CONCLUSION Conference on Electrical and Electronics Engineering, ICEEE (2004),
p.g 182-187.
We have discussed the importance of studying and 6. John Justus C ,”Predictive models of economic systems based on data
analyzing census data with the help of data mining and mining”,2015 International Workshop on Data Mining with Industrial
machine learning. The new trends in census preparation and Applications,DMIA 2015,Part of the ETyC 2015 (2016),p.g 61-65.
census data dissemination procedures and data mining and 7. Joon Heo,Okyu Kwon,”User-driven Economic Data Analysis
Framework”, 2014 IEEE 10th World Congress on Services, p.g
machine learning methodologies to analyze and predict socio 263-264.
economic status of different areas are studied. This provided 8. Sharath R , Krishna Chaitanya S, Nirupam K N, Sowmya B J and Dr K G
a brief idea about data mining and machine learning Srinivasa,”Data Analytics to predict the Income and Economic
techniques applications on census and house hold data. The Hierarchy on Census Data”,2016 International Conference on
Computational Systems and Information Systems for Sustainable
first step in this process is data cleaning, and finding highly Solutions,CSITSS 2016 p.g 249-254.
correlated attributes that represents the financial status of the 9. Octavia Juarez Espinosa, Chris Hendrickson Ph.D., James H. Garrett Jr.
people. After that k-means and k-mediods techniques were Ph.D,”Visualization of economic input-output data”,Proceedings of the
applied for clustering data to find people living below International Conference on Information Visualisation (1999),isssue
1999-January,vol 1,p.g 496-501.
poverty and above poverty. In this experiment it was found
that k-mediods is giving more accuracy, precision, recall,
f-measure and less error rate when compared to k means
algorithm.

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3812 & Sciences Publication
 Socio-Economical Status of India using Machine Learning Algorithms

10. Ying Cai, Jiangping Chen, Xiaoqing Fan, Zhenguo Yu,”Study on the
Regional Economic Data Analysis and Mining Platform Base on
OLAM”,2010 Second International Workshop on Education
Technology and Computer Science,ETCS 2010 (2010),vol 3,p.g
817-820.
11. Avery Sandborn and Ryan N. Engstrom,”Determining the Relationship
Between Census Data and Spatial Features Derived From
High-Resolution Imagery in Accra, Ghana”,IEEE Journal of Selected
Topics in Applied Earth Observations and Remote Sensing (2016),vol
9,issue 5,p.g1970-1977.
12. Ferdin Joe J,Dr. T. Ravi,John Justus C”Classification of Correlated
Subspaces Using HoVer Representation of Census Data”,2011
International Conference on Emerging Trends in Electrical and
Computer Technology, ICETECT 2011 (2011),p.g 906-911.
13. Zhuang Cheng,”Regional Economic Indicators Analysis Based on Data
Mining”,Proceedings - 2014 5th International Conference on Intelligent
Systems Design and Engineering Applications, ISDEA 2014
(2014),issue 2, p.g 726-730.
14. Vital, T. P., Lakshmi, B. G., Rekha, H. S., & DhanaLakshmi, M. (2019).
Student Performance Analysis with Using Statistical and Cluster
Studies. In Soft Computing in Data Analytics (pp. 743-757). Springer,
Singapore.

AUTHORS PROFILE

V. Balasankar completed his M.Tech in Computer

Neural Networks from Gokul Engineering College,
A.P., he has 14 years of teaching experience. He is
currently working as Associate Professor in Department
of Computer Science and Engineering, Aditya College
Of Engineering And Technology (ACOE) Surampalem
A.P, India. he is a member of CSI. He has published 4
papers in reputed international journals and 1 in conference. His main
research interests are Computer Networks, Machine Learning, Deep
Learning and GIS

Dr. Suresh Varma Penumatsa is a Professor in the

Department of Computer Science and Rector of
Adikavi Nannaya University, Rajahmundry, India. He
is having 21 years of teaching experience. His areas of
interest include communication networks, image
processing, speech processing and Machine Learning.
He has published several research papers. He is a life
member of ISTE, SMORSI, ISCA and IISA.

Published By:
Retrieval Number: E6610018520 /2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.E6610.018520 3813 & Sciences Publication