Predicting Students' Academic Performance Using Artificial Neural Networks: A Case Study

Document Sample
Predicting Students' Academic Performance Using Artificial Neural Networks: A Case Study Powered By Docstoc
					                                                        (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                  Vol. 8, No. 5, 2010

  Predicting Students' Academic Performance Using
      Artificial Neural Networks: A Case Study
                   Ghaleb A. El-Refae                                                    Qeethara Kadhim Al-Shayea
       Faculty of Economics and Admin. Sciences                            Faculty of Economics and Admin. Sciences, MIS Dep.
           Al-Zaytoonah University of Jordan                                         Al-Zaytoonah University of Jordan
                    Amman, Jordan                                                             Amman, Jordan
               ghalebalrefae@yahoo.com                                                   kit_alshayeh@yahoo.com


Abstract—Predicting students’ academic performance is                  computer orientation classes, use of computer-multimedia,
critical for universities because strategic programs can be            disposition toward computers, and majors.
planned in improving or maintaining students’ performance.
The goal of this study is to predict the factors affecting the             Al-Tamimi and Al-Shayeb [5] investigated some factors
university students' performance using Artificial Neural               affecting student performance in the fundamentals of
Networks (ANN) model. Various factors that may likely                  financial management course at United Arab Emirates
influence the performance of a student were identified.                University.
Generalized Regression Neural Network (GRNN) is used to
                                                                           Ibrahim and Rusli [6] developed three predictive models
predict the university students' performance. It is noticed a
                                                                       using SAS Enterprise Miner that are, artificial neural
significant improvement in the prediction made by
                                                                       network, decision tree and linear regression. The result of
GRNN due to its generalization property. The most                      this study showed that all of the three models produce more
important predictor variable influencing performance is                than 80% accuracy. It also showed that artificial neural
consistently having the largest regression. Results showed that        network outperforms the other two models.
secondary school performance which is measured by scores in
secondary school certificate examination, measured in a                    Oladokun, Adebanjo and Charles-Owaba [7] presented
percentage form having the largest regression value.                   an artificial neural network model for predicting the likely
                                                                       performance of a candidate being considered for admission
    Keywords-component; regression, stdudent performance;              into the university was developed and tested.
Artificial neural networks; general regression network
                                                                                   II.    ARTIFICIAL NEURAL NETWORKS
                     I.    INTRODUCTION
                                                                           An artificial neural network (ANN) is a computational
   The prediction and explanation of academic performance              model that attempts to account for the parallel nature of the
and the investigation of the factors relating to the academic          human brain. An (ANN) is a network of highly
success and persistence of students are topics of utmost               interconnecting processing elements (neurons) operating in
importance in higher education [1].                                    parallel. These elements are inspired by biological nervous
   McKenzie and Schweitzer [2] presented a study that was              systems. As in nature, the connections between elements
a prospective investigation of the academic, psychosocial,             largely determine the network function. A subgroup of
cognitive, and demographic predictors of academic                      processing element is called a layer in the network. The first
performance of first year Australian university students.              layer is the input layer and the last layer is the output layer.
                                                                       Between the input and output layer, there may be additional
    Alfan [3] determined the undergraduate students'                   layer(s) of units, called hidden layer(s). Fig. 1 represents the
performance in the Faculty of Business and Accountancy,                typical neural network. You can train a neural network to
University of Malaya and the factors influencing the                   perform a particular function by adjusting the values of the
performance of the undergraduate students. The result of the           connections (weights) between elements.
study shows that the predictor variables do explain the
variance in the students' final cumulative grade point
average. In addition, it was found that knowledge prior to
entering the university such as economics, mathematics and
accounting is crucial in assisting the students in undertaking
the courses in both business and accounting program. The
study also found that female students perform better than
male students; whilst Chinese students perform better than
Malay and Indian students.
    Su [4] evaluated the performance of university students
who learned science texts by using, information
communication technologies including animation, static
figures, power point, and e-plus software. The results
included the computation of the F-ratio, p-values, and
Cohen’s effect-sizes of attitudes toward science and learning
science in relation to the student’s gender, attendance of                               Figure 1. A typical neural network




                                                                  97                                   http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                                  (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                            Vol. 8, No. 5, 2010
    For the researcher and the financial analyst, the main
advantage of ANNs is that there is no need to specify the
                                                                                Di2 = ( x − ui ) ( x − ui ) , the squared distance between the
functional relation between variables. Since they are                                            T
connectionist-learning machines, the knowledge is directly
imbedded in a set of weights through the linking arcs among                     input vector x and the training vector u, x= the input vector,
the processing nodes. In order to train a neural network                        ui=training vector i, the center of neuron i, spread=a constant
properly one needs a large set of representative 'good                          controlling the size of the receptive region.
quality’ examples. In the case of bankruptcy problems, the
researcher should be cautious when drawing conclusions
from neural networks trained with only one or two hundred
cases, as observed in most previous studies [8].
A. Generalized Regression Neural Network
    The GRNN was applied to solve a variety of problems
like prediction, control, plant process modeling or general
mapping problems [9].
    General regression neural network Specht [10],
Nadaraya [11] and Watson [12], does not require an iterative
training procedure as in back-propagation method.
    The GRNN is used for estimation of continuous
variables, as in standard regression techniques. It is related
to the radial basis function network and is based on a                           Figure 2. Generalized Regression Neural Network (GRNN) Architecture
standard statistical technique called kernel regression. By
definition, the regression of a dependent variable y on an                                       III.   EXPERIMENTAL RESULTS
independent x estimates the most probable value for y, given
x and a training set. The regression method will produce the                    A. Data
estimated value of y, which minimizes the mean-squared                             This study was conducted at the faculty of Economics
error. GRNN is a method for estimating the joint probability                    and Administrative Sciences, Al-Zaytoonah University of
density function (pdf) of x and y, given only a training set.                   Jordan in Hashemite Kingdom of Jordan. Our sample
Because the pdf is derived from the data with no                                consists of 208 students belonging to accounting
preconceptions about its form, the system is perfectly                          department. The information for this study has been
general. Furthermore, it is consistent; that is, as the training                obtained from the register office at Al-Zaytoonah University
set size becomes large, the estimation error approaches zero,                   of Jordan and which are maintained on a computerized
with only mild restrictions on the function. In GRNN,                           database.
instead of training the weights, one simply assigns to wij the
target value directly from the training set associated with
input training vector i and component j of its corresponding                        The Cumulative Grade Average Point (CGPA) is used as
output vector [13]. GRNN architecture is given in Fig. 2.                       an indicator to measure the performance of the university
GRNN is based on the following formula [14]:                                    students'.
                                ∞                                                   The students' overall performance was hypothesized to
                                ∫ y. f ( x, y ).dy                              be a function of the following factors: (1) Secondary school
                   E[y | x] =   −∞                                              performance is measured by scores in secondary school
                                 ∞
                                                                     (1)        certificate examination, measured in a percentage form (2)
                                 ∫ f ( x, y ).dy
                                 −∞
                                                                                type of secondary school branch, (3) gender, and (4)
                                                                                boarding or non boarding student.
where y is the output of the estimator, x is the estimator                      B. Results Analysis
input vector, E[y|x] is the expected output value, given the                        A generalized regression neural network (GRNN) with a
input vector x and f(x,y) is the joint probability density                      radial basis layer and a special linear layer and linear output
function (pdf) of x and y.                                                      neurons was created using the neural network toolbox from
The function value is estimated optimally as follows:                           Matlab 7.9 as shown in Fig. 2. Generalized regression neural
                                 n                                              networks are a kind of radial basis network that is often used
                                ∑ h .w      i       ij
                                                                                for function approximation.
                         yj =   i =1
                                       n
                                                                    (2)             The first layer has as many neurons as there are input/

                                    ∑h
                                                                                target vectors. Each neuron's weighted input is the distance
                                                i                               between the input vector and its weight vector. Each
                                     i −1                                       neuron's net input is the product of its weighted input with
where wij= the target output corresponding to input training                    its bias. Each neuron's output is its net input passed through
vector xi,                                                                      radial basis transfer function. Radial basis transfer function
             − Di2                                                              is a neural transfer function which calculates a layer's output
                                                                                from its net input. If a neuron's weight vector is equal to the
hi = e   2. spread 2
                       , the output of the hidden layer neuron,                 input vector (transposed), its weighted input will be 0, its net




                                                                           98                                   http://sites.google.com/site/ijcsis/
                                                                                                                ISSN 1947-5500
                                                                 (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                           Vol. 8, No. 5, 2010
input will be 0, and its output will be 1. The second layer                      The GRNN with cumulative grade point average CGPA
also has as many neurons as input/target vectors.                            as a target and gender as input was been created. The spread
                                                                             value was chosen 0.5. The percent correctly predicted in the
     We used a spread slightly lower than the distance                       simulation sample is approximately 27 percent as shown in
between input values, in order, to get a function that fits                  Fig. 6.
individual data points fairly closely. A smaller spread would
fit data better but be less smooth.




      Figure 3. A generalized regression neural network (GRNN)

The GRNN with cumulative grade point average CGPA as a
target and secondary school performances that is measured
by scores in secondary school certificate examination,
measured in a percentage form as input was been created.
Then simulate the network with 208 inputs. The network
outputs after simulation. The spread value was chosen 0.2.                                               Figure 6.
The percent correctly predicted in the simulation sample is
approximately 76 percent as shown in Fig. 4.                                     The GRNN with cumulative grade point average CGPA
                                                                             as a target and boarding or non boarding as input was been
                                                                             created. The spread value was chosen 0.5. The percent
                                                                             correctly predicted in the simulation sample is
                                                                             approximately 20 percent as shown in Fig. 7.




                             Figure 4.

    The GRNN with cumulative grade point average CGPA
as a target and type of secondary school branch as input was
been created. The spread value was chosen 0.6. The percent                                               Figure 7.
correctly predicted in the simulation sample is
approximately 36 percent as shown in Fig. 5.




                                                                                                         Figure 8.

                             Figure 5.




                                                                        99                                 http://sites.google.com/site/ijcsis/
                                                                                                           ISSN 1947-5500
                                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                             Vol. 8, No. 5, 2010
    It is obvious from the results, that the most affecting                         Faculty of Economics and Administrative Sciences at Al-Zaytoonah
factor in academic university students' which is measured by                        University of Jordan, Amman – Jordan. His research interest is in the
                                                                                    application of IT and IS in Business and Economics.
CGPA is the secondary school performance.
    Fig. 8 shows the multiple regressions for the four
                                                                                                              Qeethara Kadhim Abdul Rahman Al-Shayea, has
affecting factors. The spread value was chosen 0.2. The                                                       received Ph. D. in Computer Science, Computer
percent correctly predicted in the simulation sample is                                                       Science Department, University of Technology,
approximately 89 percent.                                                                                     Iraq, 2005.      She received her M.Sc. degree in
                                                                                                              Computer Science, Computer Science Department
                         IV.     CONCLUSIONS                                                                  from University of Technology, Iraq, 2000. She has
                                                                                                              received her High Diploma degree in information
    In this paper the general regression neural network is                                                    Security from Computer Science Department,
used for the prediction of university student performance.                                                    University of Technology, Iraq, 1997. She has
The advantage of using the GRNN in the prediction is its                                                      received B. Sc. Degree in Computer Science
generalization property. The results of this study provide                                                    Department from University of Technology, Iraq,
evidence which suggests that secondary school performance                                                     1992. She joined in September (2001-2006),
                                                                                    Computer Science Department, University of Technology, Iraq as assistant
is the single most important variable associated with their                         professor. She joined in September 2006, Department of Management
overall performance upon graduation from university. Other                          Information Systems Faculty of Economics & Administrative Sciences Al-
variables such as type of secondary school branch, gender,                          Zaytoonah University of Jordan as assistant professor. She is interested in
and boarding or non boarding student show a lesser degree                           Artificial intelligent, image processing, computer vision, coding theory and
of significance in predicting performance as compared with                          information security.
secondary school score. The results also indicate that the
variables examined in this study provided a significant
contribution in predicting performance when used jointly
with secondary school performance variable.
                               References
[1] P. Fenollar, S. Roma´n and P. J. Cuestas, University students’ academic
performance: An integrative conceptual framework and empirical analysis,
British Journal of Educational Psychology, 77, pp. 873–891, 2007.
[2] K. Mckenzie and R. Schweitzer, Who Succeeds at University? Factors
Predicting academic performance in first year Australian university
students, Higher Education Research and Development, Vol. 20, Issue 1,
pp. 21-33, May, 2001.
[3] E. Alfan, Undergratuate Students' Performance: The Case of University
of Malaya, Quality Assurance in Education: An International Perspective,
Vol. 13, Issue 4, pp. 329-343, 2005.
[4] K. Su, An integrated science course designed with information
communication technologies to enhance university students’ learning
performance, Computers and Education, Vol. 51, Issue 3, pp. 1365-1374,
Nov., 2008.
[5] H. A. Al-Tamimi and A. R. Al-Shayeb, Factors Affecting Student
Performance in the Introductory Finance Course, Journal of Economics and
Administrative Science, Vol. 18, No. 2 Dec., 2002.
[6] Z. Ibrahim and D. Rusli, Predicting Students’ Academic Performance:
Comparing Artificial Neural Network, Decision Tree and Linear
Regression, 21st Annual SAS Malaysia Forum, 5th September 2007.
[7] V.O. Oladokun, A.T. Adebanjo, and O.E. Charles-Owaba, Predicting
Students’ Academic Performance using Artificial Neural Network: A Case
Study of an Engineering Course, The Pacific Journal of Science and
Technology, Vol. 9, No. 1, pp. 72-79, May, 2008, web page available at:
[8] J. C. Neves and A. Vieira, Improving Bankruptcy Prediction with
Hidden Layer Learning Vector Quantization, European Accounting
Review, Vol. 15, No. 2, pp. 253–271, 2006.
[9] D. W. Patterson, Artificial Neural Networks, Theory, and Applications,
Englewood Cliffs, NJ: Prentice-Hall, 1995.
[10] D. F. Specht, A general regression neural network, IEEE Trans. Neural
Networks, vol. 2, pp. 568–576, 1991.
[11] E. A. Nadaraya, On estimating regression, Theory of Probab.
Applicat., vol. 9, pp. 141–142, 1964.
[12] G. S. Watson, Smooth regression analysis, Sankhya Series A, vol. 26,
pp. 359–372, 1964.
[13] M. T. Hagan, H. B. Demuth, M. Beale, Neural network design, PWS
Publishing Company, Boston, 1996.
[14] K. Kayaer and T. Yildirim, Medical Diagnosis on Pima Indian
Diabetes Using General Regression Neural Networks, web page available
at: www.yildiz.edu.tr/~tulay/publications/Icann-Iconip2003-2.pdf.


                          AUTHORS PROFILE

Ghaleb A. El-Refae, has a Ph. D. and M.A in Financial Economics form
USA, M. Sc and B. Sc in Accounting. He is a professor and Dean of




                                                                              100                                       http://sites.google.com/site/ijcsis/
                                                                                                                        ISSN 1947-5500

				
DOCUMENT INFO
Description: IJCSIS invites authors to submit their original and unpublished work that communicates current research on information assurance and security regarding both the theoretical and methodological aspects, as well as various applications in solving real world information security problems.