Evaluating Students Performance by Artificial Neural Network using WEKA

Full text

International Journal of Computer Applications (0975 – 8887)
Volume 119 – No.23, June 2015
36
Evaluating Students Performance by Artificial Neural
Network using WEKA
Sumam Sebastian
M-Tech Computer and Information Science
College of Engineering Poonjar
Jiby J Puthiyidam
Assistant Professor
Dept. of Computer Science and Engineering
College of Engineering Poonjar
ABSTRACT
Data mining is the process of extracting hidden patterns and
useful information from large set of data is now becoming
part of current inventions. Data mining now can be applied to
different fields like marketing, education; health etc. Data
mining in field of education is named as educational data
mining. Educational data mining can help institutions to
predict the performance of their students so as to improve
their academic results. In this paper artificial neural network is
used to predict the performance of student. Multilayer
Perceptron Neural Network is used for the implementation of
prediction strategy. Experiment is conducted using weka and
real time dataset available.
Keywords
Data Mining; Educational Data Mining; Artificial Neural
Network; Multilayer Perceptron Neural Network(MLP);
Association Rule Mining;
1. INTRODUCTION
Data mining [1] is the process of analyzing data from different
perspectives and summarizing it into important information so
as to find hidden patterns from a large data set. Data mining
[2] points to the strategy of discovering of implicit, previously
unknown and practically useful information from the data in
the databases. It uses techniques of machine learning,
statistical and visualization to discover and present knowledge
in a form which is easily understandable to us. The abundance
and fast evolution of the data mining discipline comes from its
large variety of research areas of interest. Data mining
applications adopts different kind of parameters to examine
the data. Educational Data Mining[3] is a newly emerged
technique that helps to discover methods that will explore
unique types of data from education database and helps to
predict students’ academic performance.
Figure 1: Educational Data mining System
It is very necessary for an institution to maintain a good
academic result, for that student’s academic performance has
to be maintained in better manner. So a continuous student’s
performance evaluation strategy has to be invented. Different
kinds of data mining techniques can be used for this like
association rule mining.K-means clustering, artifiacial neural
network etc.Among the different methods most advanced and
accurate method is the evaluation using artificial neural
network.
2. OBJECTIVES
The main objectives are, first to determine all the personal and
academic factors that affects the performance of student,
second to transform these factors to a suitable form for system
coding and third is to model a neural network that can predict
the performance based on the data of student. The main
concept used in this paper is that of artificial neural network.
3. THE ARTIFICIAL NEURAL
NETWORKS
An artificial neural network (ANN)[4], often called as a
"neural network" (NN), is a computational model based on the
biological neural networks, in other words, is a representation
and emulation of human neural system. It consists of an
interconnected group of artificial neurons and processes
information using a connectionist approach to computation. In
practical terms neural networks are non-linear statistical data
modeling tools [5]. They can be used to model complex
relationships between inputs and outputs or to find patterns in
data. Using neural networks as a tool, data warehousing firms
are harvesting information from datasets in the process known
as data mining [6].
Multilayer Perceptron
The most popular form of neural network architecture is the
multilayer perceptron (MLP). A multilayer perceptron:
 Has any number of inputs.
 Has one or more hidden layers with any number of
units.
 Uses generally sigmoid activation functions in the
hidden layers.
 Have connections between the input layer and the
first hidden layer, between the hidden layers, and
between the last hidden layer and the output layer.
Figure 2: Feed forward neural network

International Journal of Computer Applications (0975 – 8887)
Volume 119 – No.23, June 2015
37
MLP is especially suitable for approximating a classification
function which sets the example determined by the vector
attribute values into one or more classes. MLP trained with
back propagation algorithm is used for data mining.
4. DATA COLLECTION AND
PREPROCESSING
In this research paper the data was collected from the two
classes 8 and 9 of Sacred Hearts Girls' High School
Bharananganam Kerala.A dataset of 300 students was used
for the evaluation. Neural network is used for predicting the
student performance. The attributes selected are mainly of two
types, first academic attributes related to the academic details
of student and personal attributes related to the personal
details of student that affects the study and performance of
study of student. The academic attributes selected are
1.Interest of study of the student categorized as low, average
and good, 2.Unit test mark the average mark of unit tests
conducted divided as low, average and good, 3.Assignment
mark which is the average of all assignments divided as
average and good, 4.Attendance score which is the average of
attendance of the student taken divided into average and
good,5.Extracurricular activities performance which is the
performance of student in other activities along with studies
grouped in to low average and good,6.Residence which is the
staying of student categorized into either hostler or day
scholar. The personal attributes selected are 1.parent’s
education and family status where parents education divided
as poor average and good and 2.family status is divided as low
and average and good. In a given dataset Data Pre-Processing
technique is used to identify noise data, missing attribute
values, irrelevant and redundant data.
Table1. Attributes and Its Possible Values
5. METHODOLOGY
In this research we have used Weka for the entire
implementation. Simple training and testing using multilayer
perceptron neural network was done first. For this the entire
data set was divided into two separate tests. Half for training
set and another half for testing set. Training was done by
adjusting the different learning and momentum rates. Among
the training results the best was taken for analysis. Second
MLP training after association rule mining is done.
Association rule mining extracts the important rules so that it
helps to identify the important attributes. Unnecessary
attributes are removed from
the data set and MLP neural network training is done. It gives
a better result than simple train & test method. For the most
fare evaluation of result K-fold cross validation method of
MLP training was used. In this the entire data set is not
divided into two different sets as of prior, instead as the input
to the system whole data set is given.10 filed cross validation
is used here. In this the entire data set is divided into 10
subsets. Among this one set is used as test set and remaining
nine sets are used training set. The entire evaluation steps are
depicted in the following figure
Figure 3: Work Methodology
5.1 WEKA Environment
WEKA [8] stands for Waikato Environment for Knowledge
Learning. It was developed by the University of Waikato,
New Zealand. WEKA supports many data mining tasks such
as data re-processing, classification, clustering, regression and
feature selection to name a few.
The supported data formats are ARFF, CSV, C4.5 and binary.
Alternatively you could also import from URL or an SQL
database. After loading the data, preprocessing filters could be
used for adding/removing, attributes, discretization, Sampling,
randomizing etc.Weka is a collection of machine learning
algorithms for data mining & machine learning tasks. Weka is
open source software issued under the GNU General Public
License.
5.2 MLP Training with WEKA
Two sets are used for MLP neural network training in WEKA.
They are Training set and Test set [9].
ATTRIBUTES
DESCRIPTION
VALUES
INTEREST OF
STUDY
Interest of student
in studying
Low
Average
Good
UNIT TEST MARK
Average mark of
student in unit tests
Low
Average
Good
ASSIGNMENT
Average of marks
of assignments
Average
Good
ATTENDANCE
Average
attendance of
student in the class
Average
Good
EXTRACARRICULAR
ACTIVITIES
Performance of
student in
extracurricular
activities
Low
Average
Good
RESIDENCE
Residence of
student in studying
ie in hostel or not
Non
Hoster
Hostler
PARENT
EDUCATION
Education of the
parents of student
Poor
Average
Good
FAMILY STATUS
The total family
status of the
student
Low
Average
and good

International Journal of Computer Applications (0975 – 8887)
Volume 119 – No.23, June 2015
38
 Training set: A set of examples used for learning that is
to fit the parameters [i.e., weights] of the classifier.
 Test set: A set of examples used only to assess the
performance [generalization] of a fully-specified
classifier.
Back Propagation algorithm is used for the network training.
Back Propagation algorithm
Initialize all weights to small random numbers
Until satisfied DO
 For each training example Do
1. Input the training example to the network and
compute the training outputs
2. For each output unit k
3. For each hidden unit h
4. Update each network weight
Where
Here we are adjusting the learning rate and momentum to get
a better training result.
Main parameters for learning: hiddenLayers,
learningRate,momentum,trainingTime(epoh, seed .The
parameter setting function is given as
weka.classifiers.functions.MultilayerPercep
tron -L 0.3 -M 0.2 -N 500 -V 0 -S 0 -E 20 -H
a
hiddenLayers -- This defines the hidden layers of the neural
network. This is a list of positive whole numbers. 1 for each
hidden layer. Comma seperated. To have no hidden layers put
a single 0 here. This will only be used if autobuild is set.
There are also wildcard values 'a' = (attribs + classes) / 2, 'i' =
attribs, 'o' = classes , 't' = attribs + classes.
learningRate -- The amount the weights are updated.
momentum -- Momentum applied to the weights during
updating.
For fare evaluation, the ‘cross-validation’ scheme is used
K-fold Cross Validation
 The data set is randomly divided into k subsets.
 One of the k subsets is used as the test set and the
other k-1 subsets are put together to form a training
set.
5.3 Association Rule Mining
In EDM, association rule learning is a conventional and well
researched method for determining interesting relations
between attributes in large databases [10]. Association rule
Mining is mainly intended to recognize strong rules from
databases using different measures of support and confidence.
Support (s) and confidence (c) are two measures of rule
interestingness. They truely reflect the usefulness and
certainty of the discovered rule.
Apriori Algorithm
Apriori is a seminal algorithm proposed by R. Agarwal and R.
Srikant in 1994 for mining frequent item sets for Boolean
association rules.
The following lines state the steps in generating frequent item
set in Apriori algorithm. [11]
Let Ck be a candidate item set of size k and Lk as a frequent
item set of size k. The main steps of iteration are:
 Find frequent set Lk-1
 Join step: Ck is generated by joining Lk -1 with itself
(Cartesian product Lk-1 x Lk-1)
 Prune step (apriori property): Any (k − 1) size item set
that is not frequent cannot be a subset of a frequent k size
item set, hence should be removed
 Frequent set Lk has been achieved [11].
The parameter setting function in weka for association rule
mining is
weka.associations.Apriori -N 10 -T 0 -C 0.9
-D 0.05 -U 1.0 -M 0.1 -S -1.0 -c -1
car -- If enabled class association rules are mined instead of
(general) association rules.
classIndex -- Index of the class attribute. If set to -1, the last
attribute is taken as class attribute.
delta -- Iteratively decrease support by this factor. Reduces
support until min support is reached or required number of
rules has been generated.
6. PERFORMANCE EVALUATION
To evaluate the performance of above methods of neural
network training different parameters are available like
Accuracy, Precision, Recall, F-Measure, Kappa score etc.
Here accuracy, precision and recall are considered.
 Accuracy (percent correct)
Accuracy is how close a measured value is to the actual (true)
value. Accuracy retrieves the percentage of correctly
classified instances.
Accuracy= TP+TN
TP+FP+TN+FN
 Precision
Precision is a value of the accuracy provided by a unique class
that was predicted.
Precision= TP
TP+FP
 Recall
Recall is a measure of the ability of a prediction model to
select instances of a certain class from a data set. It is c also
called sensitivity, and points to the true positive rate.
Recall= TP
TP+FN

International Journal of Computer Applications (0975 – 8887)
Volume 119 – No.23, June 2015
39
Where
TP =True Positive
TN=True Negative
FP=False Positive
FN=False Negative
7. RESULTS AND DISCUSSION
The table below shows the values obtained by various
performance evaluation parameters.
Table 2: Results of different neural network training
No.
Method of training
Learning rate
Momentum rate
Accuracy (%)
Precision
Recall
1
Simple Train &
Test
0.7
0.6
35
0.5
0.35
2
Simple Train &
Test after
Association Rule
Mining
0.9
0.8
62
0.76
0.62
3
K-Fold Cross
validation of Train
& Test
0.2
0.3
91
0.95
0.91
In this table the considered parameters are method of training,
learning rate, momentum rate, the accuracy precision and
recall obtained on that specified learning and momentum rate.
Learning and momentum rate that gives better training result
is considered here. The values given to learning rate and
momentum can range from 0.2 to 1.0.The three different
training has three different results. So a better method
selection is easier. Accuracy indicates how accurate the
training method is. Here k-fold cross validation is better than
MLP neural network training method.
8. CONCLUSION
This paper presented one use of data mining in the educational
data mining field for predicting student’s performance.
Artificial neural network was used here for prediction.k-fold
cross validation gives the most accurate result than basic
training method and training after association rule mining.
Association rule mining retrieved the most important
attributes that affects the performance of student. Those
attributes are mark of unit test, mark of assignment and
attendance in the class.MLP training with this attributes gave
a far better result than simple training.10 fold cross validation
is used here for the training. The data set considered here is
the real time dataset of marks of 300 students. As a future
work fuzzy logic can be implemented to increase the
performance evaluation result of the student.
9. REFERENCES
[1] Han J. and Kamber M.: “Data Mining: Concepts and
Techniques,” Morgan Kaufmann Publishers, San
Francisco, 2000.
[2] Anwar, M. A., and Naseer Ahmed. Knowledge Mining in
Supervised and Unsupervised ssessment Data of
Students’ Performance." 2011 2nd International
Conference on Networking and Information Technology
IPCSIT vol. Vol. 17. 2011.
[3] http://www.educationaldatamining.org/JEDM/index.php/
JEDM
[4] V.O. Oladokun, Ph.D., A.T. Adebanjo, B.Sc., and O.E.
Charles-Owaba, Ph.D. “Predicting Students’ Academic
Performance using Artificial Neural Network: A Case
Study of an Engineering Course.”
[5] Refaat, M. Data Preparation for Data Mining Using SAS,
Elsevier, 2007.
[6] S. M. Kamruzzaman and A. M. Jehad Sarkar “A New
Data Mining Scheme Using Artificial Neural
Networks”,2011.
[7] Amrender Kumar: “Artificial Neural Network for Data
Mining”.
[8] WEKA MANUAL
[9] Jung-Woo Ha.” Classification using Weka (Brain,
Computation, and Neural Learning)”.
[10] Predicting Student Performance by Using Data Mining
Methods for Classification; Dorina Kabakchieva Sofia
University “St. Kl. Ohridski”, Sofia 1000.
[11] Paresh Tanna, Dr. Yogesh Ghodasara:” Using Apriori
with WEKA for Frequent Pattern Mining”.
[12] Baha Sen, Emine Ucar. Evaluating the achievements of
computer engineering department of distance education
students with data mining methods. Procedia Technology
1 262 – 267, 2012.
IJCATM : www.ijcaonline.org