You are on page 1of 5

Neural Network Model to Predict Electrical

Students’ Academic Performance


Pauziah Mohd Arsad1, Norlida Buniyamin1, Jamalul-lail Ab Manan2
1
Engineering and Technical Education Research Group,
Faculty of Electrical Engineering, Universiti Teknologi MARA, MALAYSIA
2
Strategic Advanced Research Cluster, MIMOS Berhad, MALAYSIA
pauzia167@salam.uitm.edu.my

Abstract - Students performance is very crucial to any educational Shah Alam, Selangor offers degree and postgraduate Civil,
institution particularly in the engineering field. This paper Mechanical, Chemical and Electrical Engineering programs
describes a neural network based model (NN model) for academic running separately in its own faculty [1] [2]. Electrical
performance prediction of Electrical Engineering Degree students at Engineering diploma programmes are offered in UiTM branch
the Faculty of Electrical Engineering, Universiti Teknologi MARA
(UiTM), Malaysia. The study was conducted on student intakes from
campuses in the states of Sarawak, Terengganu and Pulau
Matriculation entry level. The performance was measured based on Pinang. A high percentage of these students upon graduation
their cumulative grade point average (CGPA) upon graduation. The will usually undertake the Bachelor of Electrical Engineering
students’ results for fundamentals subjects at first semester are used Programme offered by UiTM. They will be allowed to register
as predictor variables (initial values) for predicting the expected in the 3rd semester of the programme. The Matriculation
(projected) final CGPA upon graduation using Artificial Neural students which are considered the “normal route” will start in
Network (ANN). The outcomes of the study indicated that there the 1st semester. Matriculation students are those that
appears to be a direct correlation between students’ results for core undertook a one year matriculation programme after finishing
subjects at semester one with the final overall academic performance the Malaysian secondary school programme[1] [2].
irrespective of their gender. It can be ascertained that the analysis on
strong students’ abilities in engineering fundamentals contributed
strongly in influencing the overall academic performance in In order to graduate, students must undertake a total of 137
Engineering. Based on the outcomes of this study, we believe that credits with a minimum CGPA score of 2.00. [3] Thus,
strategic interventions can be done during their study period to students with a Diploma can graduate within six semesters or
improve their final performance, which can be extracted from this three years, since they have been given credit exemption for
prediction model. subjects in semesters 1 and 2.

Keywords - Prediction, engineering fundamentals, academic This paper presents the investigation which was conducted on
performance Neural Network the performance of Matriculation entry level students. Gender
based performance was also investigated. Further, we
developed a Neural Network Model to predict the students’
I. INTRODUCTION performances based on their entry level and academic
Students’ performance and achievement are important issues performances in core subjects in the earlier semesters. We
in any university management. Upon completion of used the final student’s CGPA to measure academic
undergraduate program, they can proceed further to pursue performance. CGPA is the easiest way of measuring the
into post-graduate studies, work in the government or private students’ performance as other traits such as inter–personality
sectors or even start entrepreneurial companies based on their skill, communication skill and other leadership skill constitute
experiences during students’ lives. Thus student performance intangible parameters. CGPA score is the famous criteria used
is of dire importance to determine future undertakings upon by many recent employers as a basis to call for first interview
graduation. Predicting students’ performances become the key session. [4]
success factor for university management because good
students’ performances mean that the university will get better II. PERFORMANCE PREDICTION
ranking amongst universities. Hence, university management Traditionally, academic researchers have developed several
must plan strategic interventions along their study periods to statistical models such as regression and decision trees, to
improve the overall achievement upon graduation. predict students’ performance. The data input was mainly
taken from questionnaires which include student’s family
Universiti Teknologi MARA (UiTM) is one of the institutes background, participation in co-curriculum and involvement
of higher learning in Malaysia that offers engineering courses. in class. [5]
The Bachelor of Engineering (Honours) Electrical Programme
evolved from the Advanced Diploma Programme which R .R Kabra [6] used Decision Tress Model to predict First
started in 1968. Currently, UiTM’s main campus based in Year Student performance based on students earlier records in
school that includes Maths, Science, and gender and parents’ form the foundation for UiTM Bachelor of Electrical
occupation. K Kardigama [7] on the other hand used Engineering course.
Multilayer Perceptron Neural Network (MPNN) to predict
students’ performance in the final examination on III. ARTIFICIAL NEURAL NETWORKS
Thermodynamics only using tests, Quiz and assignments as Artificial Neural Network (ANN) is a biological inspired
input variables. Another researcher Oladokum V O [8] intelligent technique that is generally made of a number of
predicted students’ final performance upon graduation by highly interconnected processing elements units or nodes
using Artificial Neural Network (ANN). He again used whose functionality is loosely based on animal neuron. The
demographic factors such as previous school location and processing ability of the network is stored in the inter-unit
zone, type of school, parents’ education, O level results and connection strengths, or weights, obtained by a process of
age at entry level as input variables. S Sembiring et al [9] on adaptation to, or learning from a set of training patterns.
the contrary used Data Mining Technique (DMT) to predict Nueral Network (NN) can be used to predict future events
students’ performance. They used questionnaire that cover based on historical data. Most of the early application of NN
psychometric factors such as Interest, Study Behaviour, have been in systems such as aerospace simulation,
Believe, Family Support and Engage Time as input variables Automotive, Banking, Defence, Financial, Medical and
for 1000 students. Naik B and Ragothanam S [10] again market research. In this paper, we use NN in the field of
used NN to predict MBA students’ success using education.
undergraduate information like campus location, gender,
ethnic status, CGPA and age as input variables. A ANN consists of a network of neurons and each neuron is
comparative study between NN, DT and LR was done by associated with an input vector, a weight vector corresponding
Ibrahim Z and Rusli D [11] to predict students’ academic to the input vector, a scalar bias, a transfer function and an
performance using students’ demographic profile and CGPA output vector. A neural network may consist of one or more
at first semester as predictor variables. layers of neurons with one or more neurons in each layer. The
final layer is called the output layer and all previous layers are
It can be seen that demographic profile as input variables is called hidden layers. In the hidden layers, the output of a layer
very popular among academic researchers in the past. The key becomes the input to the following layer. The transfer function
difference between them and our study is that our approach of a neuron converts the input to the output of the neuron.
focuses on only students’ abilities at first semester, Multilayer neural networks are quite powerful tools used in
irrespective of their family background or previous academic solving many different problems compared to single layer. In
history once accepted into the specific Degree Program. this paper Multilayer Back propagation neural network is
Instead of using questionnaires, we used live data of students’ used. [13, 14]
performance based on three generations of Matriculation
intakes. We hope that our study can be used by university The network is designed using the fastest technique so as to
management and students who joined Degree Program in get the minimum error between targeted and predicted values.
Engineering to use our methodology, and model to help them Usually, more number of layers and more neurons in each
complete the entire Program successfully. layer can produce a better network with minimum error rate
between targeted and predicted. Thus, designing a neural
Data from Matriculation students from intakes in July 2005, network is a trial and error process of choosing the Learning
2006 and 2007 were collected and compiled for this study. Rate (0 -1) and the Momentum rate (0-1). The Learning rate
The data included gender, subjects attempted and Grade Point helps to accelerate the convergence of ANN training process
scored for every subject, CGPA at first semester and the while the Momentum rate helps to accelerate the training
CGPA scored at semester eight. There were 391 students’ data process. There are many techniques in ANN and in this study
collected and 7 subjects attempted which include Lavenberg-Marquardt (LM) was chosen, with two hidden
Mathematics, Signal and System, Circuit Theory, layers and one output layer. The activation or transfer function
Fundamental Electronics, Communication Theory, used in hidden layers are Hyperbolic Tangent known as
Laboratory1 and Co-curriculum. However, for this study, we “logsig” while the output layer is mainly pure linear or
omitted Laboratory and Co-curriculum as they do not form the “purelin”. The training will stop when Minimum Squared
pre-prequisite to the other subjects in the higher semesters. Error (MSE) is obtained, that is the minimum error between
We have concluded from our earlier study that the final CGPA targeted and predicted by NN model. The output only has one
performance was independent of gender. Hence, in this study, output, which is the CGPA at semester eight. The NN model
we have also assumed the same. Female students can perform was built using the NN toolbox of the MATLAB software
equally well upon graduation provided that they have strong version R2010a. The model was trained and tested with 391
analytical capabilities right from semester one. [12] data for Matriculation students.

In this paper, our study mainly focusses on Electrical Degree


Students and the corresponding fundamental subjects that
TABLE 1.0: INPUTS AND OUTPUT OF NN IV. RESULTS AND ANALYSIS
Fig. 1 and 2 below show the results that we obtained for
Matriculation students. Fig. 1.0 shows the Performance at
minimum Mean Squared Error (MSE) for the Matriculation
students. From Fig. 1.0, it can be seen that the best validation
performance was at MSE= 0.0544 while the predicted CGPA8
is shown in Fig. 2.0.

From Fig. 2.0, we observed that there is a slight variation


between Predicted and Targeted CGPA8. It can be seen that at
lower CGPA8, the predicted is higher than the targeted
CGPA8 while those high CGPA8 was predicted lower. We
believe that the variations have shown a few points; firstly,
that the model is able to predict the expected result of the
students’ performance in the final semester, secondly, that the
students have improved, throughout the duration of their study,
with the help of lecturers intervention, through mentoring and
In this study, firstly the data from past years students’ coaching. The graph also shows, the cultural shock that the
performance were collected based on subjects in the first student undergo when the first joined the university, hence
semester for Matriculation intakes. The data include the final resulting in worse than predicted performance in early
CGPA at semester eight as the target output for the NN semesters.
training. The input predictor variables are Circuit Theory,
Fundamental Electronics, Signal and System, Mathematics,
Communication Theory and Grade Point Average of 1 st
semester. The target output was CGPA8. Table 1.0 shows the
inputs and output of the model. As shown, two models were
developed, namely Model A (Table 1.0) which considers the
five different subjects attempted and Model B, which
considers the most predicted variables as inputs. Table 2.0
depicts the developed NN Prediction models A and B.

TABLE 2.0: NN PREDICTION MODELS

Fig. 1.0 Best Validation with MSE=0.05544


Fig. 3.0 Best Validation with MSE =0.0933
Fig. 2.0 Comparison Between
Targeted and Predicted for Tested Data (6 inputs)
V DISCUSSIONS AND RECOMMENDATION
The same technique was used with three input predictor In this section, we describe the practical steps taken to do
variables that seemed to be the most important subjects in intervention for the Engineering Degree students.
semester one namely Circuit Theory 1, Fundamental Strategic steps include taking lesser total credits for the
electronics and Signals and Systems 1 or Model B. Best medium abilities or lower CGPA scorers at semester one and
performance for the Matriculation Model B is shown in Fig. they tend to extend before they complete the total credit hours
3.0 below. required. The weaker students can take up courses during
intersession or summer schools to reduce the total credit hours
From the graphs of Target and Predicted CGPA8, it can be (load) during normal terms. Such students can only take
seen that similar results as the previous model were obtained. incremental courses one at a time. The role of academic
We observed again, those with lower CGPA have been advisors is very important at this time around as students need
improved, as shown in Fig 4.0 which is higher than the the proper advice and motivation to keep steadily strong until
predicted CGPA8 in both cases for the Matriculation students. the very end of their studies.
Hence, strategic intervention and actions can be encouraged
by the Academic advisors so as to improve the overall CGPA
during the period of studies as they have additional seven (7)
more semesters before reaching final semester. As those with
high CGPA at the start, they must stay focus and maintain
their performance as the predicted is lower than the targeted
due to increasing credit hours as they moved to higher
semesters in the Program.

Fig. 4.0 Comparison between Targeted and Predicted


for Tested Data (3 inputs).
As for those who scored high CGPA at the start of the course, [2] Arsad P M, Buniyamin N, Manan J A, “Proposed Academic Students’
Performance Prediction Model: A Malaysian Case Study” ICEED, 3rd
they have to work smart so as to maintain the same CGPA, as
International Congress on Engineering Education 2011.
seen from those graphs that the predicted is much less than the [3] Arsad P M, Buniyamin N, Rasheid N A (2010)”Evaluation of
actual CGPA8. It is extremely difficult to maintain CGPA as Academic Performance of Electrical Engineering bachelor Degree
the total credit hours keep on increasing every semester and Students from Various Entry Levels: a Case Study”, 9th WSEAS
International Conference on Education and Educational Technology
the subjects are getting more difficult as far as electrical
(EDU ’10)
courses are concerned. The subjects at semester one only form [4] Zaharim, A et al, “Perception and Expectation Towards Engineering
the basis foundation or perquisite to other subjects like Graduates by Employer: a Malaysia Study Case”, WSEAS
Circuit Theory 2, Electromagnetic Theory, Electrical Transaction on Advances in Engineering Education-Issue 9 Volume 6
September 2009
Machines, Control Systems, Microprocessor and Electronics
[5] N Ali et al, “The factors Influencing Students Performance at
Design. The students need stay focus and maintain their Universiti Teknologi MARA Kedah, Malaysia” Management Science
interest as well determination all along until graduation. and Engineering, Vol 3 No 4 2009
[6] Kabra R.R, Bichkar R.S “Performance Prediction of Engineering
Students Using Decision Trees” International Journal of Computer
Application, Vol. 36, No. 11 2011
VI CONCLUSION [7] Kadirgama Ket al, “Development of Genetic Algorithms and
In this paper, we presented a neural network based model (NN Multilayer Perceptron Neural network (MPNN) Model to study The
model) to predict the academic performance of Electrical Students Performance in Thermodynamics” International Engineering
Education Conference, Medinah 2009.
Engineering Degree students at the Faculty of Electrical
[8] Oladokun V.O “Predicting Students’ Academic Performance using
Engineering, Universiti Teknologi MARA (UiTM), Malaysia. Artificial Neural Network: A Case Study of an Engineering Course”
The important parameters namely, the fundamental subjects The Pacific Journal of Science and Technology, Volume 9. Number 1.
which the student take in the earlier semester are taken as May-June 2008 (Spring)
[9] Sembiring S,et al “Prediction of Student Academic Performance by an
predictor variables. In our study, subjects such as Signals and
Application of Data Mining Technique” International Conference on
Systems, Fundamental Electronics and Circuit Theory were Management and artificial Intelligence 2011
considered as strong predictor parameters. The final semester [10] Naik B, Ragothaman S. “Using Neural Network to Predict MBA
eight CGPA score is the output of the prediction model. Student Success” College Student Journal 2004, Vol.38 No.1
[11] Ibrahim Z, Rusli Daliela, “Predicting Students’ Performance:
Experimental data obtained, were used for training and testing
Comparing Neural Network, Decision Tree and Linear Regression”
the network. Several criteria such as training algorithm, 21st Annual SAS Malaysia Forum 2007
learning rate, momentum coefficient, number of hidden layer, [12] Arsad P M et al, “Female Electrical Engineering Degree Students
number of neurons in each hidden layer and activation Performance Based on Different Entry Levels: A Malaysian Case
Study” WSEAS 2011
function were given to improve the performance of the ANN.
[13] I Musirin, “Short Course on ANN and Evolutionary Programming”
2007
We have shown that there is a direct correlation between [14] Active Media, “Applying NN with MATLAB Hands-On Training”
students’ results for core subjects at semester one with the 2006
[15] Essentials of Statistics for the Behavioral Sciences 4th Edition
final overall academic performance irrespective of their
Frederick J Gravetter
gender. It can be ascertained that the analysis on strong [16] ISIS: UiTM Information System Management
students’ abilities in engineering fundamentals contributed
strongly in influencing the overall academic performance in
Engineering. Based on the outcomes of this study, we believe
that strategic interventions can be undertaken during their
study period to improve their final performance, which can be
extracted from this prediction model. The developed model
could be used as evaluation tool by the Head of Programs or
academic advisors to enable them to play active roles with
the goal to improve students’ academic performance.

ACKNOWLEDGMENT
The authors gratefully acknowledge and thank University
Teknologi Mara (UiTM), Shah Alam, Malaysia, for the
Research Intensive Faculty (RIF) Grant (600-
RMI/DANA/5/3/RIF (195/2012).

REFERENCES

[1] Arsad P M, Buniyamin N, Manan J A, “Profiling the Performance of


Electrical Engineering Bachelor Degree Students Based on Different
Entry Levels”, International Journal of Education and Information
Technologies Issue 2, Volume 5, 2011

You might also like