APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN ESTIMATING PARTICIPATION IN ELECTIONS

International Journal of Information Technology, Modeling and Computing (IJITMC) Vol.1, No.3,August 2013
DOI : 10.5121/ijitmc.2013.1303 23
APPLICATION OF ARTIFICIAL NEURAL NETWORKS
IN ESTIMATING PARTICIPATION IN ELECTIONS
Seyyed Reza Khaze1
, Mohammad Masdari2
and Sohrab Hojjatkhah3
1
Department of Computer Engineering, Dehdasht Branch, Islamic Azad University, Iran,
khaze@iaudehdasht.ac.ir, khaze.reza@gmail.com
2
Department of Computer Engineering, Urmia Branch, Islamic Azad University, Iran,
m.masdari@iaurmia.ac.ir
3
Department of Computer Engineering, Dehdasht Branch, Islamic Azad University, Iran,
hojjatkhah@gmail.com
ABSTRACT
It is approved that artificial neural networks can be considerable effective in anticipating and analyzing
flows in which traditional methods and statics aren’t able to solve. In this article, by using two-layer feed-
forward network with tan-sigmoid transmission function in input and output layers, we can anticipate
participation rate of public in kohgiloye and Boyerahmad Province in future presidential election of
Islamic Republic of Iran with 91% accuracy. The assessment standards of participation such as confusion
matrix and ROC diagrams have been approved our claims.
KEYWORDS
Anticipating, Data Mining, Artificial Neural Network, political behaviour, elections
1. INTRODUCTION
Anticipating the political behavior of the public will demonstrate the future landscape of a society
in which everyone can experience the future space virtually and begin to program and make
policies about his/her job career, political, economic, and military and other sections affairs being
informed about available characteristics, parties and candidates viewpoints. As many political
theorists have a lot of problems in anticipating public political behavior in these kinds of elections
and aren’t able to do it accurately using statistical tools and social and political analyzing methods
[1], so, data mining methods such as artificial neural networks are used to find potential rules in
these kinds of data by gathering information and necessary documents of people previous
electoral and political behaviors [2].
Artificial neural network is considered as one of the modern mathematical-computational
methods which are used to solve un-anticipated dynamic problems in developed behavioral
systems during a time period. By learning to recognize patterns from data in which other
computational and statistical method failed to solve them, artificial neural networks are able to
solve the problems. Nowadays, artificial neural networks are widely used due to their ability to
anticipate. Considering that this feature , participation anticipate and election results are among
the most interesting challenges for political scientists, it can be used artificial neural network to
anticipate in political issues, although it was less used in Political Science by scholars. Political

24
scientists and scholars are already search for vast models to anticipate election results throughout
developing world.
They use survey models which are a series of indicators to construct analytical model of electoral
behavior, local election analyses and previous election results survey which isn’t possible to
anticipate accurately due to possible nature problems and uncertainty of new electoral behaviors
and results [3]. Artificial neural networks can be seen as an approximation of the actual neural
network because it simulates how biological neurons act in the human brain structure in which the
same processing units of neurons are connected to each other by synapses. These networks are a
set of interconnected processing units that analyze the model in which a set of input data mapped
out into a set of output units. The artificial neural networks are used to anticipate and classify [4].
In the first part of this paper reviews the applications of artificial neural networks to predict. In
the next section, the Estimating Participation in Elections results using these networks have been
discussed and then participate in the election of the case study is discussed. This is done using a
two-layer feed-forward network. And finally discuss the evaluation of the performance of
artificial neural networks.
2. PREDICT USING NEURAL NETWORKS
The increasing tendency to anticipate using artificial neural networks resulted in a remarkable
increase in research activities in the recent decade [5]. Artificial neural networks are suitable
methods to indicate their efficiency in anticipating exchange rate, analyzing economical time
series, issues related to stock and stock market. Results and the function of artificial neural
networks have shown that this model has better function than popular methods considering
assessment criteria of future time sequences anticipation. In recent years, many complicated
statistical methods have been developed and used for anticipation process in the related issues.
However, there are two basic problems about these methods. It includes personal statistical
problem, power and certain analyses for a single and multi-dimensional time series. Artificial
neural networks were reliable methods to remove statistical problems in anticipating multi-
dimensional time series [6].
In the economic field, it is hard to anticipate macro-economic issues due to the lack of accurate
and satisfying model. The most accurate model for economical anticipating is the time series
model of black-box model which partly considers the economic structure. High noise levels, short
time series, and nonlinear effects are introduced as time series problems solutions of regression
method which are the popular ways to solve economical anticipations in the recent years. The
given solution of artificial neural networks can be used to solve noted problems to anticipate
macro-economic issues. By selecting the advanced parameters in artificial networks, the
researchers have been provided the experimental results to anticipate the indicator of industrial
production in USA in which the given results of their simulation experiments indicate the better
function of artificial neural networks than popular linear time series and regression [7].
In political field, the flexible model of neural networks is capable of recognizing and anticipating
the empirical relationship between democracy and global collision theory while the logical model
is unable to do so. The logic model was the proposed model of Marchi et al. They claimed it is
superior than artificial neural network which rejected by Nathaniel beck et al. They began to
basically analyze Grynaviski, Gelpi & Marchi theory and emphasized that their theories are
completely in common with theirs about the world. Their common viewpoint is using main
standards to assess the world studies in order to anticipate out of sample functions. They also
indicated that other conclusions of Marchi et al are incorrect. They approved that using same
assessment criterion for each two models which Marchi et al had been used was based on logic

25
model superiority to the other neural networks which provided challenges and rejected their
theory [8].
In tourism field, artificial neural network model had better function to anticipate international
tourism demands than other popular anticipation methods, time series and regression techniques.
It is approved in a research executed by Rob Law & Norman Au. The authors have been used
organized Feed-Forward network which is a particular model of neural network to anticipate
Japanese trip to Hong-Kong. Networks consisting of input and output layers in which input layer
includes 6 service price nodes, average hotel rates, exchange rates, population, internal gross
expenditures and output layer of a node is to represent a travel demands from Japan to Hong -
Kong. The achieved experimental results obtained from the study of the time interval from 1967
to 1996 has shown that the neural network model has better results than other methods [9].
In Environmental area, neural network can be used to solve available problems and issues of
processes modeling of real world by designing composed architecture with meta-heuristic
algorithms. These problems include nonlinear multidimensional space and chaos theory.
Although, designing neural networks for high efficiency is difficult, but Harry Niska et al have
been used multi-layer perceptron model to anticipate Nitrogen Dioxide during day time in a
crowded traffic station in Helsinki city by parallel combining of genetic algorithm in parallel to
choose input and design high level architecture. It removed the practical problems of designing
neural networks for high efficiency and indicated that combined networks can guarantee it [10].
In informational technology field, neural networks have also shown that short-term intervals can
provide better function than adapted methods of time series to anticipate traffic rate in top-in
networks. Paulo Cortez et al have been executed several experiments to assess anticipation
accuracy in the noted methods using real-world data and related information to internet service
providers. The experiments have been done in separated time sequences of 5 min, 1 h and a day.
However, time series of 1-day anticipation had better results but the neural group experiments
have shown that it can be provided best results for short-term time series [11].
3. ESTIMATING PARTICIPATION IN ELECTIONS USING NEURAL NETWORK
Multilayer Feed-Forward is an example of artificial neural networks in which learning of a neural
network is performed by back propagation algorithm. It frequently repeats the weight learning
process to classify and anticipate of class from tuples. A Multilayer Feed-Forward network
consists of an input layer, one or more hidden layer and output layer. Each layer consists of units.
The input network is in the size of features which have educational samples and enters to the
input layer, simultaneously. These inputs are transported through input layer and then become
weighted. Then, they are transported to the second layer of pseudo-neuron units which known as
hidden layer. The output of hidden layer can be entered to another hidden layer as an input of
output layer. The weighted output or hidden layers are used as input of output layer in which they
can reach to the ideal weight and network learning by educational data series and provide
classification and anticipation operation for tuples and test samples [12].We suppose to anticipate
participation rate in 11th election of Islamic Republic of Iran in Kohkiloye and Boyerahmad
Province using artificial neural network. These data includes features and prospects of 100
qualified persons to attend in election [1] which is shown its fields in Tables 1-9.
Table 1.Age of people
LabelID
Old1

26
Middle-aged2
Young3
Table 2.Degree of people
LabelID
PhD1
Masters2
BA3
Under license4
Table 3.Jobs of People
LabelID
Government employees1
Professors and university staff2
Free jobs3
Table 4.Political Orientation of people
LabelID
Reformists1
Fundamentalists2
"Velayi"3
Table 5.Opinion of people about government services
LabelID
Mortgage1
targeted subsidies2
Marriage Loans3
Fuel Rationing4
Table 6: Opinion of people about Participation type in elections
LabelID
Religious duty and religious1
Participation in decision making2
Overall reform3
Table 7.Opinion of people about general policy in international affairs
LabelID
Resistance1
Negotiation2
Concessions3
Table 8.Opinion of people about the election officials
LabelID
Higher accuracy1
Confidence2
Unreliability3

27
Table 9.Opinion of people about Candidates
LabelID
Popular candidate1
Party candidate2
Political elite3
Table 10.Opinion of people about Participation in elections
LabelID
Partnership1
possible participation2
Without participation3
We, firstly, create an artificial neural network using MATLAB tools. Our artificial neural
network is a Feed-Forward network with tan-sigmoid transmission function in the hidden and
output layers. In this network, we use 10 neurons in hidden layer. The network has 10 inputs and
3 outputs as the target vector has 3 members.
Figure 1.Overall architecture of the Feed-Forward Neural Network to Predict Participation
We provide the number of hidden layers equal to 10 and create it based on 10 hidden layers. In
pre-process stage, the data of similar additional rows are removed if it is available. In the third
stage, we classify all input data to groups, randomly and use some data for educational sample,
assessment and the remains for network test. The number and percentage of each stage are shown
in Table 11.
Table 11.number and percentage of each stage
Stage Sample
Training 70
Validation 15
Testing 15
4. DISCUSSION
In network learning part, we begin to learn it by choosing Gradient learning function. It is
necessary to note that Mean square error function is selected to recognize network efficiency. To
find performed anticipation rate and real values, it is used error function concept. It is also used
mean square error to assess anticipation error [13, 14]. If the assessment error repeated 6
consecutive times, the learning procedure must be stopped. In this case, it is occurred in 15

28
repetitions. The diagram of learning errors, assessment errors and test errors are shown in Fig
(2).
Figure 2.Diagram of training, evaluation and testing error of Feed-Forward Neural Network to Predict
Participation
It is clear that the best efficiency is occurred in repetition 9. The learning function of learning data
is shown in repetition 15 in Fig 3
Figure 3.Training Performance of Feed forward Neural Networks to Predict Participation
In Fig 4, the histogram of learning stage errors are indicated.

29
Figure 4.Histogram of errors in training
In assessments, it is also used confusion matrix which indicates the classified samples of a class
in another class or similar one [15]. The reaction of distribution matrix analyses and different
errors about the final network are indicated in Fig 5
Figure 5.Collision Matrix to predict participation

30
As it can be seen, the used neural network to anticipate participation result in election can provide
91% accuracy in future election of Islamic Republic of Iran in Kohkiloye and Boyerahmad
Province. The ROC curve also indicates the real positive rate and false positive rate. False
positive rate is a proportion of negative tuples which recognize positive wrongly and provide for a
model [13, 14]. Diagonal elements in each table indicate the number of cases which classified
correctly. The non-diagonal elements are those which classified wrongly. The blue-colored cell
(in the lower right corner of the table) indicates the total wrong classification percent rate in red
color. In ROC curve, the positive accurate percent rate is outlined opposite of wrong positive
percent rate
Figure 6.ROC curve for prediction of participation
As it can be seen in ROC diagrams, neural network has good classification function.
5. CONCLUSIONS
Artificial neural network is considered as one of the modern mathematical-computational
methods which are used to solve un-anticipated dynamic problems in developed behavioral
systems during a time period. Nowadays, artificial neural networks are widely used due to their
ability to anticipate. Considering that this feature , participation anticipate and election results are
among the most interesting challenges for political scientists, it can be used artificial neural
network to anticipate in political issues. We use artificial neural networks to predict participation
in presidential elections in Iran. Our data consist of 100 individual attitudes of Dehdasht
citizenship. We use a neural network feed forward neural network with tan-sigmoid transfer
function. We have been able to participate in the election with 91 percent accuracy to predict.
ROC graphs and other standards of our claim are proved.
9. REFERENCES
[1] Sangar, A. B., Khaze, S. R., & Ebrahimi, L (2013), Participation Anticipating In Elections Using Data
Mining Methods , International Journal On Cybernetics & Informatics,2(2),47-60.

31
[2] Gharehchopogh, F. S. & Khaze, S. R. (2012). Data Mining Application for Cyber Space Users
Tendency In Blog Writing: A Case Study, International journal of computer applications, 47(18), 40-
46.
[3] Gill, G. S. (2008), Election Result Forecasting Using Two Layer Perceptron Network ,Journal of
Theoretical and Applied Information Technology, 47(11),1019-1024
[4] Caleiro, A. (2005), How to Classify a Government? Can a Neural Network do it? , University of
Evora, Economics Working Papers.
[5] Adya, M., & Collopy, F. (1998), How Effective Are Neural Networks at Forecasting and Prediction?
A Review and Evaluation, J. Forecasting, 17, 481-495.
[6] Thiesing, F.M., Vornberger, O. (1997), Sales Forecasting Using Neural Networks, Neural Networks,
International Conference On ,4, 2125-2128 .
[7] Moody, J. (1995), Economic Forecasting: Challenges and Neural Network Solutions, In Proceedings
of the International Symposium on Artificial Neural Networks
[8] Beck, N., King, G., & Zeng, L. (2004), Theory and Evidence in International Conflict: A Response to
De Marchi, Gelpi, and Grynaviski, American Political Science Review, 98(2), 379-389.
[9] Law, R., & Au, N. (1999), A Neural Network Model To Forecast Japanese Demand For Travel To
Hong Kong, Tourism Management, 20(1), 89-97.
[10] Niska, H., Hiltunen, T., Karppinen, A., Ruuskanen, J., & Kolehmainen, M. (2004), Evolving The
Neural Network Model For Forecasting Air Pollution Time Series, Engineering Applications of
Artificial Intelligence, 17(2), 159-167.
[11] Cortez, P., Rio, M., Rocha, M., & Sousa, P. (2012), Multi‐Scale Internet Traffic Forecasting Using
Neural Networks And Time Series Methods, Expert Systems, 29(2), 143-155.
[12] Towell, G. G., & Shavlik, J. W. (1994), Knowledge-Based Artificial Neural Networks, Artificial
intelligence, 70(1), 119-165.
[13] Gorunescu, F. (2011), Data Mining: Concepts, Models and Techniques (Vol. 12), Springerverlag
Berlin Heidelberg.
[14] Han, J., Kamber, M., & Pei, J. (2006), Data Mining: Concepts And Techniques. Morgan Kaufmann.
[15] Yang, S., Yan, J., Gao, C., & Tan, G. (2011), Blogger’s Interest Mining Based On Chinese Text
Classification, In Nonlinear Mathematics for Uncertainty and its Applications (pp. 611-618), Springer
Berlin Heidelberg.
Authors
Seyyed Reza Khaze is a Lecturer and Member of the Research Committee of the
Department of Computer Engineering, Dehdasht Branch, Islamic Azad University, Iran.
He is a Member of Editorial Board and Review Board in Several International Journals
and National Conferences. His interested research areas are in the Software Cost
Estimation, Machine learning, Data Mining, Optimization and Artificial Intelligence.
Mohammad Masdari is Currently Head of the Department of Computer Engineering,
Urmia Branch, Islamic Azad University, Iran. He Has a Currently PhD Candidate In
Department Of Computer Engineering At Science And Research Branch, Islamic Azad
University, Iran. His Interested Research Areas Are in the Wireless Sensor Networks,
data mining, optimization and artificial intelligence.
Sohrab Hojjatkhah is Currently Head of the Department of Computer Engineering,
Dehdasht Branch, Islamic Azad University, Iran. He Has a Bachelor's Degree In Software
Engineering, Received From Amir Kabir University, Iran, Then Received A Master's
Degree In Artificial Intelligence From The University Of Shiraz And Currently PhD
Candidate In Department Of Computer Engineering At Science And Research Branch,
Islamic Azad University, Iran. His Interested Research Areas Are In The Image
Processing, Speech Processing, Machine Learning, Data Mining And Artificial
Intelligence.

APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN ESTIMATING PARTICIPATION IN ELECTIONS

More Related Content

What's hot (20)

Viewers also liked (14)

Similar to APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN ESTIMATING PARTICIPATION IN ELECTIONS (20)

Recently uploaded (20)

APPLICATION OF ARTIFICIAL NEURAL NETWORKS IN ESTIMATING PARTICIPATION IN ELECTIONS