PREDICTING STUDENTS’ PERFORMANCE IN

PALAWAN STATE UNIVERSITY – ROXAS CAMPUS USING NEURAL NETWORK

PREDICTIVE ANALYTICS

1. Introduction

Philippines has been ranks a poor seventh among the nine Southeast Asian nations in the area of
education and innovations according to co-chairman of National Competitive Council (NCC) and also
among the countries with highest school dropout rate. The dropout rates revealed an alarming 83.7%,
meaning that the country is producing 2.13 million college dropouts annually. In this case, the Philippine
government must, in the next education generation must be able to reverse the current situation from 80
percent of college students enrolled in private schools and 20 percent in state universities and colleges
(SUCs) to 20 percent, private colleges, and 80 percent SUCs (Manila Bulletin, 2012).

A significant problem in higher education in Philippines is the poor results of the students after the
admission. Because of this, many students leave universities leave after the first year of study. According
to the study of Oanca, Dragoescu and Ciucu (2013), some of the reasons for these results are: poor
background knowledge in the field of study, very low grades and incapacity of passing an examination,
lack of financial resources and the increasing cost of education. That’s why one of aims of the educational
institutions is the development of strategies to prevent schools dropouts and it should be among the top
priorities to avoid the phenomenon of early school leaving. Since the ratio between graduating students
and enrolled students is one of the indicators used for school accreditation of higher education institutions
in Philippines. The universities’ management tries to keep this ration at a good value by means of early
intervention and supporting those student with problems and those candidates to leave school. With this,
predicting students’ result is an important problem and has a great value to universities for early intervention
to avoid early school leaving.

Many studies and researches tried to predict the students’ results based on various data. Predictions
were made using different statistical method like multivariate regression and path analysis. But none of
these methods have the power of discovering potential patterns as neural networks. The feed forward neural
networks have been applied in many fields like medical diagnosis, stock forecasting, and bankruptcy
prediction and classification purposes because they are one of the best functional mappers. The good results
of applying neural networks in classification problem lead me to use it for predicting students’ results in
higher education. (Oanca, Dragoescu & Ciucu, 2013).

The purpose of this study is to develop a feed forward neural network prediction tool used to predict
the GPA after the first year of study in Palawan State University. It will also reviews the related work in
the field using neural network for prediction and classification purposes in education.

1.1 Neural Networks

Neural Network (Artificial Neural Network) – the common name for mathematical structures and
their software or hardware models, performing calculations or processing of signals through the rows of
elements, called artificial neurons, performing a basic operation of your entrance. The original structure
was inspired by the natural structure of neurons and neural systems, particularly the brain.

In addition, artificial neural networks are the modeling of the human brain with the simplest
definition and building blocks are neurons. There are about 100 billion neurons in the human brain. Each
neuron has a connection point between 1,000 and 100,000. In the human brain, information is stored in
such a way as to be distributed, and we can extract more than one piece of this information when necessary
from our memory in parallel. We are not mistaken when we say that a human brain is made up of thousands
of very, very powerful parallel processors. In multi-layer artificial neural networks, there are also neurons
placed in a similar manner to the human brain. Each neuron is connected to other neurons with certain
coefficients. During training, information is distributed to these connection points so that the network is
learned.

Moreover, artificial neural networks is an algorithms that simply mimic the way our brain works.
They consists of an array of interconnected nodes that exchange information among each other (Gerritsen,
Leon, 2017), comparable to the way our neurons, connected by dendrites and axons, exchange information.
They learn iteratively over time by observing different examples, similarly to how children can learn skills
from their parents by observation. However, unlike children that can learn recognize and object after only
observing it once. Neural networks often require a greater set of observation to attain sufficient predictive
capacity, as they are notoriously data hungry (Cerny & Proximity, 2001; Karpathy et. al., 2014). Neural
networks differ from other classification algorithms in the fact that internally, information is processed in
a parallel way, comparable to how our brain functions. This differs from the serial processing that many
other algorithms like Decision Tree classifiers use (Maren, Harston & Pap, 2014).

Figure 1: Layers of Artificial Neural Network

As shown in Figure 1 above, a neural network consists of three layers: an input layer, an
intermediate (hidden) layer and an output layer. The blue boxes shown here represent the neurons and the
arrows represent the connection points. The data set prepared for training at the input layer is shown to the
network. The network assigns the weights of the events it learns to the connection points in the intermediate
layer. Not every point has to be a value, and some points can be zero. A threshold value is added between
the layers so that the zero values at the connection points do not become zero.
The property that makes Neural Networks interesting for complex domains like computer vision,
playing complex games like GO or understanding human speech, is their ability to derive answers from
complex and imprecise data (Srivastava, Hinton, Krizhevsky, Sutskever, & Salakhutdinov, 2014). Neural
Networks are able to detect patterns that are too complicated for humans or other machine learning
algorithms to pick up. And have been used successfully in numerous business applications for pattern
recognition, prediction and classification (Gaur, 2012). These properties make them particularly suited for
a domain like EDM where large quantities of data are available and where the data is often noisy and not
linearly separable owing to the fact that it is based on human behavior.

The advantages of neural network is that they can work fine in case of incomplete information.
They do not require knowledge of the algorithm solving the problem (automatic learning). It can process
information in a highly parallel way. They can generalize (generalize to cases unknown). They are resistant
to partial damage. They can perform associative memory (associative - like working memory in humans)
as opposed to addressable memory (typical for classical computers).
According to the study of Mijwel, Maad M. (2018), the following are some advantages of ANN:

 Storing information on the entire network: Information such as in traditional programming is stored
on the entire network, not on a database. The disappearance of a few pieces of information in one
place does not prevent the network from functioning.

 Ability to work with incomplete knowledge: After ANN training, the data may produce output even
with incomplete information. The loss of performance here depends on the importance of the
missing information.

 Having fault tolerance: Corruption of one or more cells of ANN does not prevent it from generating
output. This feature makes the networks fault tolerant.

 Having a distributed memory: In order for ANN to be able to learn, it is necessary to determine the
examples and to teach the network according to the desired output by showing these examples to
the network. The network's success is directly proportional to the selected instances, and if the
event cannot be shown to the network in all its aspects, the network can produce false output.

 Gradual corruption: A network slows over time and undergoes relative degradation. The network
problem does not immediately corrode immediately.

 Ability to make machine learning: Artificial neural networks learn events and make decisions by
commenting on similar events.

 Parallel processing capability: Artificial neural networks have numerical strength that can perform
more than one job at the same time.

Neural Networks also have certain disadvantages. It is at present impossible to derive how a
network came to a certain output. Compared to a Decision Tree classifier in which one can follow a set of
steps to arrive at a classification, a Neural Network doesn’t store any explicit representation of the way it
achieved its result. As such, a Neural Network is often referred to as a black box where information goes
in, and a result comes out (Srivastava et al., 2014). They might therefore not be suited in some cases where
the decision process needs to be explicit. This would be the case with expert system giving medical advice
for which the decision process needs to be checkable. In our case, this black box nature is not a deal breaker,
as the consequence of mislabeling someone is not critical and will in most cases not require explicit
explanation (Parasuraman, Sheridan, & Wickens, 2000). Another drawback, is that Neural Networks
require vast amounts of data to achieve satisfactory performance and are therefore not suitable for every
dataset (Cerny & Proximity, 2001; Karpathy et al., 2014). However, given the size of the dataset used in
this study, with records of approximately 1600 students, this will most likely not be an obstacle.

1.2 Neural Networks in Student Performance Prediction (Related Work)

Neural networks were used by many researchers for predictions of students’ results. In (Cooper,
2010) the author presents a neural network-based decision support system that identifies students who are
“at-risk” of not retaining to their second year of study. The system correctly predicted retention for
approximately 70% of the students. (Halachev, 2012) presents a neural network used for prediction of the
outcome indicators of e- Learning, based on Balanced ScoreCard. The author obtained a 3-4% prognosis
error which is acceptable from a practical point of view. (Livieris et al., 2012) describes the implementation
of a user-friendly software tool based on neural network classifiers for predicting the student's performance
in the course of "Mathematics" of the first year of Lyceum.

Neural networks were also used in (Naik et al., 2004) to predict MBA students’ success. The
authors classified applicants to MBA program into successful and marginal students pools based on
undergraduate GPA, undergraduate major, age and GMAT score using a neural network with three layers.
They obtained a prediction accuracy for their model of about 89%. To assess the ability of the neural
network for classification of students, the authors compared the results obtained using the neural network
with a logit and probit regression model. The overall rate of prediction accuracy of the logit model was
about 73% while the accuracy of the probit was 73.37%.

(Oladokun et al., 2008) used a neural network in order to determine those factors that influence
students’ performance. They classified students in three classes according to their results. The accuracy of
prediction obtained by the authors of the paper was about 74%. (Karamouzis et al., 2008) used a three-layer
perceptron network trained by backpropagation to predict student graduation rate. The network model build
by the authors had an accuracy of 70.27% for successful graduates and 66.29% for unsuccessful graduates.

1.3 Application of Neural Networks

Neural networks have seen an explosion of internet over the last few years and are being
successfully applied across an extraordinary range of problem domains, in areas as diverse as finance,
medicine, engineering, geology, physics and biology. The excitement stems from the fact that these
networks are attempts to model the capabilities of the human brain. From a statistical perspective neural
networks are interesting because of their potential use in prediction and classification problems (Jha, Girish
Kumar)

Neural networks have been used for a wide variety of applications where statistical methods are traditionally
employed. They have been used in classification problems, such as identifying underwater sonar currents,
recognizing speech, and predicting the secondary structure of globular proteins. In time-series applications,
NNs have been used in predicting stock market performance. As statisticians or users of statistics, these
problems are normally solved through classical statistical methods such as discriminant analysis, logistic
regression, Bayes analysis, multiple regression, and ARIMA time series models (Jha, Girish Kumar).
Which indicates that the neural networks is indeed a powerful tool for data analysis.