151
Prediction of academic performance using data
mining in rst year students of peruvian university
abstract
Academic performance is a subject that has
been studied for a long time. First year students
in universities are the most vulnerable to face
performance problems, resulting in possible
desertion. Data mining in education applies data
mining techniques in the information generated in
the education sector. e present research consists of
making the prediction of the academic performance
of the students who entered the Professional School
of Computer and Systems Engineering of the
University of San Martín de Porres in the rst cycle
using data mining. Data were extracted from 1304
entrants who were classied using three factors: social,
economic and academic, and predictions were made
using three techniques: linear regression, decision tree
and support vector machines, having the best result
of 82.87% obtained using the decision tree. Out of
the dierent factors, those that most inuenced the
academic performance were the following: admission
exam grade, gender, age, income and distance from
home to the study center. Using data mining it was
possible to elaborate predictions of the academic
performance of the students, which allowed the
detection of students who could encounter issues in
their studies during the rst semester.
Key words: Academic Performance, prediction,
Educational Data Mining, EDM, Higher Education
resumen
El rendimiento académico es un tema estudiado
desde hace mucho tiempo. Los alumnos ingresantes
de las universidades son los más vulnerables a
enfrentar problemas de rendimiento, resultando en
posible deserción. La minería de datos en educación
aplica técnicas de minería de datos en la información
generada en el sector educación. El presente trabajo
consiste en realizar la predicción del rendimiento
académico de los alumnos que ingresaron a la
Escuela Profesional de Ingeniería de Computación y
Sistemas de la Universidad de San Martín de Porres
en el primer ciclo utilizando minería de datos. Se
extrajeron datos de 1304 ingresantes que fueron
E Y
L C S
R C P
V J H H
1
Universidad de San Martín de Porres.
Lima - Perú
eyamao@usmp.pe
2 Universidad de San Martín de Porres.
Lima - Perú
lcelis@usmp.pe
3 Universidad Nacional Federico Villarreal.
Lima - Perú
rcampos@unfv.edu.pe
4 Universidad de San Martín de Porres.
Lima - Perú
vhuancash@usmp.pe
Predicción del rendimiento académico mediante minería de datos en
estudiantes del primer ciclo en una universidad peruana
Recibido: junio 22 de 2018 | Revisado: julio 26 de 2018 | Aceptado: agosto 02 de 2018
https://doi.org/10.24265/campus.2018.v23n26.05
| C | L, | V. XXIII | N. 26 | PP. - | - | | -
152
clasicados en tres factores: sociales, económicos y
académicos y se realizaron predicciones a través de
tres técnicas: regresión lineal, árbol de decisiones y
support vector machines, y el mejor resultado de
82.87% se obtuvo utilizando árbol de decisiones.
De los diferentes factores, los que más inuyeron en
el rendimiento académico fueron los siguientes: nota
de examen de admisión, género, edad, modalidad
de ingreso y distancia desde su casa hasta el centro
de estudios. Utilizando minería de datos fue posible
realizar predicciones del rendimiento académico de los
ingresantes. Esto permitió la detección de ingresantes
que podrían enfrentarse a problemas en sus estudios.
Palabras clave: rendimiento académico, predicción,
educational data mining, educación superior
Introduction
One of the main challenges in
education is the exponential growth of
data generated by information systems
and technology and its use to improve
the quality of educational services
oered with a better decision making.
Educational Data Mining is an emerging
discipline that aims to take advantage of
the new capabilities of data processing and
the maturity of data mining algorithms
to enhance the learning process and
transform existing information into
knowledge. (Han, Kamber & Pei, 2012)
(Romero & Ventura, 2012) (Chalaris,
Gritzalis, Maragoudakis, Sgouropoulou,
y Tsolakidis, 2014) (Romero, Ventura,
Pechenizkiy, y Baker, 2011).
e main subject of study in higher
education is the academic performance
of students. It´s not only a tangible value
to measure the progress of a student in a
given course or subject but it´s also one
of the feature to see the level of success
during and after obtaining a degree. Is
of great importance for the educational
institutions given that the level of
success of their students is a reection
of the quality of the institution. (Calisir,
Basak, y Comertoglu, 2016) (Rodríguez
y Arenas2016) (López Bonilla, López
Bonilla, Serra, y Ribeiro, 2015) (York
2015) (Shahiri y Husain, 2015).
Prediction is one of the oldest
applications of EDM. Multiple studies
have successfully created models to
predict academic performance. It is a
complex process given that multiple
elements have been attributed to impact
academic performance of students. A
proper prediction model can be used to
detect those who might face diculties
in their studies and be in risk of dropping
out. Measures to help those students in
risk by additional tutoring, changes and
improvements in courses or curricula
are some of the adjustments made
previously. (Ramesh, parkavi y ramar,
2013) (Mishra, Kumar y Gupta, 2014)
(Strecht, Cruz, Soares, Merdes-Moreria y
Abren 2015) (ElGamal, 2013).
e most vulnerable students in
universities, who might face diculties
and drop out are found in rst year
students. Adaptation to life in a university
can be a great challenge for many and
| C | V. XXIII | N. 26 | - | 2018 |
E Y - L C S - R C P - V J H H
153
some never manage to adapt completely.
Multiple studies have found that the
number of students with failing grades
are higher in the rst year, which extends
their time in the university and in some
cases dropping out. Student retention and
their graduation are an important goal for
universities, especially in STEM majors
where student dropout rate might exceed
the 30% in rst year students. (Baradwaj
& Pal, 2012) (Sepehrian, 2012) (Li, Rusk
y Song, 2013) (Cheewaprakobkit, 2013).
Previous studies of prediction using
EDM include Mishra et al. (2014) to
predict academic performance of third
year student in computer science major
using J48 and Random Tree algorithm.
Performance in previous semesters and
other courses, leadership and motivation
are inuential in academic performance.
Elakia & Aarthi (2014) uses student
characteristics in high school to predict
in which major they will have the best
performance and discipline in universities
based on behavioral trends in high school.
Pal (2012), predicts low achievement and
chances of dropout using multiple decision
tree algorithm. Ramesh et al. (2013) uses
prediction in nal exam grade to nd those
who could fail a course, being occupation
of the parents a strong impact in the
results. Gray, McGuinness y Owende
(2014) predicted academic achievement of
rst year student, using gender, age, high
school grades, personality, motivation and
learning style. Students under 21 years
had a better prediction results. Sembiring,
Zarlis, Hartama, Ramliana, y Wani
(2011) uses SVM ad clustering to predict
academic performance using interest,
beliefs, family support, attitude and time
spent studying achieving a high prediction
rate of 93.67%.
ere is very little studies made in
Peru related to academic performance
and almost none using EDM. is study
is one of the rst to create a prediction
model from student characteristics found
in Peruvian universities.
Materials and method
Methodology
e present study used a quantitative
method. Scientic studies and related
papers where searched to determine the
proper variables needed to achieve the
objectives of this study and to create
a theoretical relationship between the
variables. e design of this study is
correlational – causal, aimed to describe
the relationships between academic
achievement and characteristics of
rst year students used to validate the
prediction methods.
Population and sample
e population of this study are the
rst year students of the Information
Systems career of the San Martin de
Porres University. e sample has been
taken from the years 2010 to 2015, giving
1304 students who were admitted, from
which transfer students (who did not take
the courses of the rst year), those who
dropped out in the rst weeks or never
enrolled in any course were removed
from the data.
Dataset
All available data from admissions
oce and faculties has been collected
and cleaned to create the dataset. From
the revision of the literature and the
| C | V. XXIII | N. 26 | - | 2018 |
P
154
data found in the data repositories of the
university, the selected variables for the
study are:
• AGE. It´s the age of the student at
the date of the rst class in university.
• GENDER. e gender, male or
female of the student, for this study
a binary representation, zero (0) for
female and one (1) for male is used.
• PROVINCE. To identify those
students that came from another
province to study in the capital.
• SCHOOLTYPE. Dierent types of
school (national, private, religious
and others).
• COLEEXC. Based on a list of the
top 500 schools with student having
the best grades in university.
• ADMISSIONEXAM. Grade of
the student in the admission exam.
Presented in percentage of the total
score.
• DISTANCE. e total distance
calculated from the place of residence
of the student to the faculty where
the classes were given. In some cases,
there was no data for the address so
the average distance from the selected
district is used.
• APPROVED. To indicate the PASS
or FAIL status of the student in the
rst year of university.
Data Mining Methods
ree data mining methods has been
selected to be applied in the dataset.
Regression: e purpose for this model
is to t the data to a model based on
variables. Answers questions like: what is
the forecast of sales for the next month?
(Han et al. 2012)
Decision Trees: Represent a group of
classication rules in shape of a tree,
based on an if-then ruleset. (Han et al.
2012).
For this study C5.0 algorithm was
selected.
Support Vector Machines: Support
vector machines is a type of algorithm
that builds a model to represent simple
point in a higher dimension to dene a
hyperplane to be used to create an optimal
separation between classes to achieve
proper classication. For the denition
of the hyperplane, the algorithm uses
the support vectors to map the data in
a high enough dimension to make the
classication. (Lantz, 2013)
Ethical aspects
e main ethical aspects about this type
of research is the privacy of the personal
data about students and professors.
Information that can be used to identify
a person requires authorization before its
use and publication.
To assure the privacy of the data
used in this study an anonymization
process was applied to the data related
to students and professors. A unique
ID was assigned to each row of the data
and every column that could be used for
individual identication like rst name,
last name, ID card number, address has
been removed. is way, protecting the
privacy and the validity of the study
results is achieved.
| C | V. XXIII | N. 26 | - | 2018 |
E Y - L C S - R C P - V J H H
155
Result and discussion
Results
Regression
Linear Regression models aim to t
into a linear model y=f(x) all the attributes
of the database don the relation between
the dependent variable (y) and the
independent variable (x). For this study,
a logistic regression model has been used
to create a model to predict the PASS/
FAIL condition of the rst year students.
For the selection of the most inuential
variables to be added to the model,
backward selection method has been
used. e variables with the best t are
shown in Table 1. As expected admission
exam score is the most important
variable and the distance has a negative
coecient, meaning that student with
less travel distance from their homes
are more likely to pass the courses. e
gender also has a negative coecient,
indicating that female student are more
likely to earn a passing grade.
Table 1
Results from logistic regression
Coecients:
Estimate Std. Error Z value Pr(>|z|) Signicance
(Intercept) -3.37717 0.87575 -3.856 0.000115 ***
ADMISSIONEXAM 2.70438 0.47570 5.685 1.31e-08 ***
AGE 0.10228 0.04379 2.335 0.019525 *
GENDER -0.55567 0.24382 -2.279 0.022663 *
DISTANCE -0.03678 0.01665 -2.209 0.027194 *
To use this model for prediction, the
dataset has been split randomly into
75% for the training and 25% for the
testing. e results of the prediction used
this model is as seen on Table 2. Further
examination of the results from this
model has shown that there is a signicant
dierence in the rate of passing the
courses on the type of admission exam
taken.
Table 2
Prediction results logistic regression
Exactude Sensibility Specicity AUC
67.4% 69.44% 59.45% 68.82%
Considering this new discovery, the
type of admission exam was used to
perform a further analysis and a new
model with a better prediction capability
was discovered using the variables in Table
3. In this new model, only considering
those who were admitted via the ordinary
type of admission process is considered.
e results from the prediction using this
model is as seen on Table 4.
| C | V. XXIII | N. 26 | - | 2018 |
P
156
Table 3
Results from logistic regression ltered by ordinary admission type
Coecientes:
Esmate Std. Error Z value Pr(>|z|) Signicance
(Intercept) -4.2345 0.5179 -8.176 2.96e-16 ***
ADMISSIONEXAM 10.9480 1.3225 8.278 < 2e-16 ***
GENDER -0.8137 0.3398 -2.395 0.0166 *
COLEEXC -1.1600 0.5839 -1.987 0.0470 *
Table 4
Prediction results logistic regression ltered by ordinary admission type
Exactude Sensibility Specicity AUC
74.4% 74% 76% 82.12%
Decision trees
e C5.0 algorithm was used to create
models based on an IF THEN rules to
study the importance of each variable
and predict the PASS / FAIL outcome.
Boosting techniques has been applied to
make multiple iterations of the algorithm
to enhance the results. Using what was
learned from the regression model, the
type of admission was used a criteria to
study the data. e best model found
using the decision trees method is as seen
on Figure 1.
Figure 1. Decision Tree model ltered by ordinary admission type
Admission exam score, gender, age
and variables related to the type of
school are considered in this model. As
in the regression model, a high admission
exam score and being female are strong
indicators of having a PASS outcome.
Prediction using this model achieved an
exactitude in the prediction of 82.87%.
| C | V. XXIII | N. 26 | - | 2018 |
E Y - L C S - R C P - V J H H
157
Support Vector Machines SVM
Support vector machines method
was used to create models based support
vectors, to create a hyperplane to make
the classication. An important part of
SVM is the selection of the kernel to be
used to model the data. e linear and
polynomial kernel couldn’t create a model
good enough to be used for prediction,
but the Gaussian and sigmoid kernels
managed to create models, achieving the
best results in predicting the outcome
the Gaussian kernel with a exactitude of
75.2%
Discussion
Prediction of the outcome of the rst
year students has been achieved. e
prediction results of the best models for
each DM method used in this study is as
seen on Table 5.
Table 5
Results of exactitude of prediction from Logistic regression, C5.0 decision trees and SVM
algorithms
Algorithm Predicon
Logisc Regression 74.4%
C5.0 82.87%
SVM 75.2%
As in other studies (Ecklund, 2013) (Li,
Swaminathan, & Tang, 2009) (Veenstra,
Dey, & Herrin, 2008) (Honken &
Ralstron, 2013) (Elakia & Aarthi, 2014)
(Gray et al. 2014), admission exam
score is an important evaluation tool for
prediction. e higher the admission
exam score, the more likely for the rst
year student to pass the rst semester.
Gender was also found to be important,
as female students had a higher chance of
passing the rst semester and the closer
the distance between the student’s place
of residence and the university, the more
likely it is to have a passing grade.
e type of elementary and junior
high school attended before studying
in the university had little impact on
student outcome.
Conclusions
Predicting the future outcome of
rst year students is an important way
for universities to detect those who will
most likely face problems to pass the
rst semester of studies, to give them the
necessary support and prevent student
dropout.
Female students in this study were
found to have a higher passing rate than
their female counterpart. Usually, being
a gender minority, as is usual in STEM
careers is considered to have a negative
impact (Bayer, Bydzovska, Geryk,
Obsıvac y Popelinsky, 2012) (Ecklund,
2013), which is a topic for future studies.
e type of school in elementary and
junior high school did not have a strong
| C | V. XXIII | N. 26 | - | 2018 |
P
158
impact in the outcome, and the students
coming from schools that are considered
to the better ones either. It is most likely
that other characteristics like emotional
maturity, motivation, group of friends
and others have a similar or stronger
impact than academic ones.
As it is often mentioned, the data
quality and quantity is important in
References
these type of studies. Given that this
study took place in one career in one
university, it is of interest for future
studies to compare the results with
other careers like accounting (Byrne
& Flood, 2008) or engineering (Li et
al., 2009) (Veenstra, Dey, & Herrin,
2008) (Cheewaprakobkit, 2013) (Li, et
al. 2013), and in other universities in
Peru.
Baradwaj, B. K., & Pal, S. (2012).
Mining educational data to analyze
students’ performance. arXiv
preprint arXiv:1201.3417.
Bayer, J., Bydzovská, H., Géryk, J.,
Obsivac, T., & Popelinsky, L.
(2012). Predicting Drop-Out
from Social Behaviour of Students.
International Educational Data
Mining Society.
Byrne, M., & Flood, B. (2008).
Examining the relationships
among background variables and
academic performance of rst year
accounting students at an Irish
University. Journal of Accounting
Education, 26(4), 202-212.
Calisir, F., Basak, E., & Comertoglu,
S. (2016). Predicting academic
performance of master’s students in
engineering management. College
Student Journal, 50(4), 501-513.
Chalaris, M., Gritzalis, S., Maragoudakis,
M., Sgouropoulou, C., &
Tsolakidis, A. (2014). Improving
quality of educational processes
providing new knowledge using
data mining techniques. Procedia-
Social and Behavioral Sciences, 147,
390-397.
Cheewaprakobkit, P. (2013). Study of
Factors Analysis Aecting Academic
Achievement of Undergraduate
Students in International Program.
In Proceedings of the International
MultiConference of Engineers and
Computer Scientists (Vol. 1, pp. 13-
15).
Ecklund, A. P. (2013) Enhancing
Incoming Male Student Retention:
An Analysis of the Experiences of
Persistence in Engineering.
Elakia, G., & Aarthi, N. J. (2014).
Application of data mining in
educational database for predicting
behavioural patterns of the
students. International Journal of
Computer Science and Information
Technologies (IJCSIT), 5(3), 4649-
4652.
ElGamal, A.F. (2013). An Educational
Data Mining Model for
| C | V. XXIII | N. 26 | - | 2018 |
E Y - L C S - R C P - V J H H
159
Predicting Student Performance
in Programming Course.
International Journal of Computer
Applications, 70(17).
Gray, G., McGuinness, C., & Owende,
P. (2014). An application of
classication models to predict
learner progression in tertiary
education. In Advance Computing
Conference (IACC), 2014 IEEE
International(pp. 549-554). IEEE.
Han, J., Kamber, M., & Pei, J. (2012).
Data Mining: Concepts and
Techniques, Elsevier.
Honken, N. B., & Ralston, P. A.
(2013). High-Achieving High
School Students and Not So
High-Achieving College Students
A Look at Lack of Self-Control,
Academic Ability, and Performance
in College. Journal of Advanced
Academics, 24(2), 108-124.
Lantz, B. (2013). Machine learning with
R. Packt Publishing Ltd.
Li, Q., Swaminathan, H. and Tang,
J. (2009), Development of
a Classication System for
Engineering Student Characteristics
Aecting College Enrollment and
Retention. Journal of Engineering
Education, 98: 361–376. doi:
10.1002/j.2168-9830.2009.
tb01033.x
Li, K. F., Rusk, D., & Song, F. (2013).
Predicting student academic
performance. In Complex,
Intelligent, and Software Intensive
Systems (CISIS), 2013 Seventh
International Conference on (pp. 27-
33). IEEE.
López
Bonilla, J. M., López Bonilla,
L. M., Serra, F., & Ribeiro, C.
(2015). Relación entre actitudes
hacia la actividad física y el
deporte y rendimiento académico
de los estudiantes universitarios
españoles y portugueses. Revista
iberoamericana de psicología del
ejercicio y el deporte, 10(2), 275-284.
Mishra T., Kumar D. & Gupta S.
(2014) Mining Students’ Data
for Prediction Performance
Fourth International Conference
on Advanced Computing &
Communication Technologies,
Rohtak, pp. 255-262.
Ramesh, V., Parkavi, P., & Ramar,
K. (2013). Predicting student
performance: a statistical and data
mining approach. International
journal of computer
applications,63(8).
Rodríguez, Á. P. A., & Arenas, D. A. M.
(2016). Programas de intervención
para Estudiantes Universitarios
con bajo rendimiento académico.
Informes Psicológicos, 16(1), 13-34.
Romero, C., Ventura, S., Pechenizkiy,
M., & Baker, R. S. (Eds.). (2011).
Handbook of educational data
mining. CRC Press.
Romero, C., & Ventura, S. (2013).
Data mining in education. Wiley
Interdisciplinary Reviews: Data
Mining and Knowledge Discovery,
3(1), 12-27.
| C | V. XXIII | N. 26 | - | 2018 |
P
160
Sembiring, S., Zarlis, M., Hartama, D.,
Ramliana, S., & Wani, E. (2011).
Prediction of student academic
performance by an application
of data mining techniques. In
International Conference on
Management and Articial
Intelligence IPEDR (Vol. 6, pp.
110-114).
Sepehrian, F. (2012). Emotional
Intelligence as a predictor of
academic performance in university.
Journal of Educational Sciences and
Psychology, 2(2).
Shahiri, A. M., & Husain, W. (2015).
A review on predicting student’s
performance using data mining
techniques. Procedia Computer
Science,72, 414-422.
Strecht, P., Cruz, L., Soares, C., Merdes-
Moreria, J. & Abren, R. (2015). A
Comparative Study of Classication
and Regression Algorithms for
Modelling Students’ Academic
Performance. In 8th International
Conference on Educational Data
Mining, Madrid, Spain, 392-395.
Veenstra, C. P., Dey, E. L., & Herrin,
G. D. (2008). Is Modeling of
Freshman Engineering Success
Dierent from Modeling of Non
Engineering Success?. Journal of
Engineering Education, 97(4),
467-479.
York, T. T., Gibson, C., & Rankin, S.
(2015). Dening and measuring
academic success. Practical
Assessment, Research & Evaluation,
20(5), 2.
| C | V. XXIII | N. 26 | - | 2018 |
E Y - L C S - R C P - V J H H
