See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352489180

Applying machine learning approach to predict students’ performance in

higher educational institutions

Article in Kybernetes · June 2021

DOI: 10.1108/K-12-2020-0865

CITATIONS READS

68 3,761

2 authors:

Nas Yakubu A. Mohammed Abubakar

Arden University Antalya Bilim University
12 PUBLICATIONS 699 CITATIONS 74 PUBLICATIONS 4,449 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by A. Mohammed Abubakar on 22 June 2021.

The user has requested enhancement of the downloaded file.

 The current issue and full text archive of this journal is available on Emerald Insight at:
https://www.emerald.com/insight/0368-492X.htm

Applying
Applying machine learning machine
approach to predict students’ learning
approach
performance in higher
educational institutions
Mohammed Nasiru Yakubu Received 13 December 2020
Revised 4 February 2021
American University of Nigeria, Yola, Nigeria, and Accepted 23 March 2021

A. Mohammed Abubakar
Antalya Bilim Universitesi, Antalya, Turkey

Abstract
Purpose – Academic success and failure are relevant lifelines for economic success in the knowledge-based
economy. The purpose of this paper is to predict the propensity of students’ academic performance using
early detection indicators (i.e. age, gender, high school exam scores, region, CGPA) to allow for timely and
efficient remediation.
Design/methodology/approach – A machine learning approach was used to develop a model based on
secondary data obtained from students’ information system in a Nigerian university.
Findings – Results revealed that age is not a predictor for academic success (high CGPA); female students
are 1.2 times more likely to have high CGPA compared to their male counterparts; students with high JAMB
scores are more likely to achieve academic success, high CGPA and vice versa; students from affluent and
developed regions are more likely to achieve academic success, high CGPA and vice versa; and students in
Years 3 and 4 are more likely to achieve academic success, high CGPA.
Originality/value – This predictive model serves as a classifier and useful strategy to mitigate failure,
promote success and better manage resources in tertiary institutions.
Keywords Information systems, Education, ICT, Artificial intelligence, Academic success,
Machine learning, Logistic regression, Enrollment data, Higher education, Nigeria
Paper type Research paper

1. Introduction
Developments in educational technologies have resulted in the accumulation of large
amounts of data on students and their learning activities. The exploration of these data has
empowered researchers and educators with knowledge and intelligence to support teaching
and learning activities (Sedrakyan et al., 2020) and to determine academic outcomes (Al-
Sudani and Palaniappan, 2019). A common indicator of academic outcome in higher
education institutions is the cumulative grade point average (CGPA). Higher education
institutions require their students to maintain a minimum CGPA of 2.0 on a 4.0 scale, to
enable them progress in their studies. High CGPA (i.e. above 2.0) are considered as academic
success that is accompanied with rewards, and low CGPA (i.e. less than 2.0) results in a
series of penalties ranging from a warning, probation, restriction, suspension or dismissal.

Kybernetes
This research did not receive any specific grant from funding agencies in the public, commercial or © Emerald Publishing Limited
0368-492X
not-for-profit sectors. The authors have no conflict of interest. DOI 10.1108/K-12-2020-0865
K There are two streams of research on the factors that contribute to the academic success
of students in higher educational institutions. The former, learning analytics are based on
data mainly derived from the use of learning management systems (LMS) applications
(Huang et al., 2020). Variables used in learning analytics include clickstream data, forum
posts, login frequency, course duration, time spent, exercises and peer interaction
(Mwalumbwe and Mtebe, 2017; Moreno-Marcos et al., 2020). The latter focuses on socio-
demographic variables such as enrollment data (Rizvi et al., 2019; Aulck et al., 2019).
Students’ enrollment information is mostly obtained at registration time such as age, gender,
ethnicity, prior education, poverty levels and parents’ qualification (Aulck et al., 2019;
Hoffait and Schyns, 2017; Kamal and Ahuja, 2019).
Research found that self-reported survey data designed to predict academic success lacks
objectivity in comparison to LMS data (Tempelaar et al., 2020). Thus, academic advising and
counseling staff in higher educational institutions rely on the CGPA scores and LMS data to
predict students’ future grades and detect those needing guidance or at risk of falling
(Gaševic et al., 2016; Shum and Ferguson, 2012). This approach is vital as it allows for the
implementation of intervention strategies (Olaya et al., 2020). However, this process captures
failing students a little too late. Similar arguments were echoed in Tempelaar et al. (2015)
influential work, which found that entry test data and demographic variables are more
robust compared to LMS data in determining student academic success. Cooper et al. (2001)
argued that qualitative and quantitative intervention strategies are needed to improve
professional practice in education.
The deficiencies of LMS data in predicting academic success has been acknowledged in
past studies (Conijn et al., 2016). For instance, LMS data cannot address course differences
and complexity. Consequently, Francis et al. (2020) stated that structural disadvantage does
impact student outcomes, particularly for disadvantaged students and this impact cannot be
identified in LMS data. Scholars recommend the use of alternative data sources (Conijn et al.,
2016; Francis et al., 2020; Tempelaar et al., 2015) and machine learning approach in
determining students’ academic success (Rincon-Flores et al., 2020). In response to these
calls, this paper opts for socio-demographic data and machine learning approach to
determine students’ academic success.
Because understanding the factors that contribute towards academic failure via an early
detection system would be more beneficial as timely intervention, in the form of guidance, it
can be made prior to students selecting courses at the beginning of a new session. In
addition, universities in the Sub-Saharan region are under resourced coupled with lack of
technologies to capture data (Bawack and Kamdjoug, 2020). This has contributed to limited
understanding of and research from the region. Few universities in the region are now
adopting technology to facilitate learning and improve their processes.
This paper’s aim is to fill the gap in the literature and also provide insights for
administrators in the region. The rest of this study is arranged as follows: the next section
presents the literature review. This is followed by the methodology section. The discussion
and results section summarize the findings, the take home lesson and ends with limitations,
and future research directions.

2. Literature review
There are numerous benefits that accompany the prediction of university students’
academic success. The most significant of these benefits is to improve the success rates of
their students. According to Alyahyan and Düs tegör (2020), university administrators rely
on the early detection of students who are at-risk, as an institution’s students’ success is an
important metric used to measure the institution’s performance. Another identified benefit
of predicting students’ success is the reduction of student attrition. Student attrition is the Applying
decrease in student numbers (dropouts)with the passage of time. Student attrition represents machine
a misuse of resources (Berens et al., 2018). These include public and private resources in the
form of money and time.
learning
In addition, Berens et al. (2018) identified the “feelings of inadequacy”, which leads to approach
social stigmatization of the individual. In countries where there is a high competition for
spaces in university programs, there is the possibility that students who dropout would
have wasted the space that could have been allocated to a student who could have fared
better. This is another result of student attrition, which could be avoided if the detection of
students’ risk is done prior to acceptance into the university. The early prediction of
academic success is also useful in allocating support and intervention procedures to at-risk
students, this helps university administrators in the planning and allocation of resources
required for the guidance and support of such students.
Data on the students is usually required to predict the success of students, and there are
several statistical techniques used in the prediction analysis. Universities collect large
amounts of data from students before they are admitted (enrollment data) and also while
they are enrolled. The former data type, enrollment data, is usually collected to gather
information about the students and is often used to process admission. In Nigerian
universities, the admission office uses this type of data to decide the programs to be offered
to applicants. For example, criteria such as the region where the student is from and their
previous academic records are used to offer programs to students. A typical academic record
used to offer admission is the Joint Admission and Matriculation Board (JAMB) examination
score.
JAMB is responsible for conducting matriculation examinations for entry into all
Nigerian universities, polytechnics and colleges of education. All university programs have
a cut-off mark, for instance, the minimum cut off mark for entry into federal universities is
between 180 and 200 points, out of a possible 400 points, depending on the university. These
cut-off marks do not only depend on the admitting university. Other variables are also
considered such as the program of study and the catchment area (region) for the university
(Kanyip, 2013). Programs such as Medicine (MBBS) and Law (LLB) require significantly
higher JAMB scores to be admitted and students in the catchment areas are favored for
admission as their score requirements are lower than students outside of the catchment area.
There is an abundance of studies on the prediction of academic success, which is based
on enrollment data in developed countries, for instance, Kovacic (2010) identified ethnicity,
course program and course block as significant variables that differentiate a successful from
an unsuccessful student. More recently, Saa et al. (2019) identified gender, nationality, high
school attended and student performance before joining the university as predictors of
student academic performance using random forest algorithm. Similar studies that have
employed demographic or enrollment data for prediction analysis, predictor variables such
as high school scores (Garg, 2018; Mohamed and Waguih, 2017), gender (Garg, 2018;
Putpuek et al., 2018), age (Hamoud et al., 2018) and ethnicity (Ahmad et al., 2015) have been
observed to determine the academic success of students. According to Francis et al. (2020),
social-economic and cultural features such as age, gender, race or ethnicity and other
personal characteristics can exacerbate differential student outcomes.
The other type of data collected from students pertains to their learning behavior while
enrolled in school is referred to as learning analytics. Learning analytics is the analysis of
data that captures a student’s learning behavior for the purpose of understanding and
improving the learning process, experience and environment. This is mainly captured via
surveys admitted to students or derived LMS. Factors such as login frequency (Hamoud
K et al., 2018; Mwalumbwe and Mtebe, 2017), discussion posts (Hamoud et al., 2018;
Mwalumbwe and Mtebe, 2017), number of downloads (Mwalumbwe and Mtebe, 2017) and
peer interaction (Mwalumbwe and Mtebe, 2017) have all been observed to be predictors of
academic success.
Traditional techniques are not suitable for analyzing large data sets (big data) obtained
from student LMS (Akgül, 2018, 2019; Baleanu et al., 2020). Sin and Muthu (2015) identified
four suitable machine learning-based techniques for educational data, namely, logistic
regression, nearest neighbor, clustering and classification. Regression is a statistical process
that involves estimating the relationship between independent variables (predictors) and
dependent variables (outcome), regression is a suitable technique for prediction and
forecasting (Kumari and Yadav, 2018). Nearest neighbour analysis is a measure of the
spread of a variable over a geographical point and groups similar observation based on the
distance to the point (Lee et al., 2019). Clustering categorizes similar objects into groups
called clusters where objects in a cluster have similar attributes and are different to objects
in other clusters (Musumeci et al., 2018). Classification identifies and assigns a certain value
to a group based on previously categorized values (Soofi and Awan, 2017).
Alyahyan and Düs tegör (2020) suggest that regression and classification are typical
examples of prediction models while clustering and association are used for descriptive
models. Prediction models use supervised learning algorithms to estimate the values
expected from the dependent variables based on the independent variables (Bramer, 2016).
Several authors have tried to identify the best educational data mining technique (regression
and classification) with regards to student prediction and the results have been inconclusive.
For instance, Rusli et al. (2008) observed that the Neuro-fuzzy model was better than
artificial neural network (ANN) and logistic regression, while Shahiri and Wahidah (2015)
found ANN to be more accurate than SVM, decision tree and Naive Bayes methods. ANN
has superior performance to regression analysis for prediction problem; however, ANN
performance for classification problem is the same with those of logistic regression and
discriminant analysis based on student data (Paliwal and Kumar, 2009).
Yaacob et al. (2019) compared five models in trying to predict student success and found
the Naives-Bayes performed better, closely followed by the logistic regression model.
Regression has also been observed to perform better than neural networks (Oyedeji et al.,
2020). In reviewing the “best practices in predicting academic success in higher education”,
Alyahyan and Düs tegör, (2020) mentioned prior-academic achievement, and student
demographics as the two most used variables in 69% of the papers they reviewed. The
authors also identified the “CGPA as the most common factors used to predict student
performance” (Alyahyan and Düs tegör, 2020). Building on this line of argument, this study
uses socio-demographic data obtained from active students (i.e. currently enrolled) as the
input variables and CGPA as the target variable. Based on the extant literature and
discussions, the following research question is proposed:

RQ. What variables (i.e. entry age, gender, state of origin, JAMB score and level of
study) are relevant for predicting Sub-Saharan students’ academic success?

3. Methodology
3.1 Data preprocessing stage
The proposed associations in this study were modelled using machine learning approach,
specifically, logistic regression. The input and target variables are enrollment data
captured by the student information system of American University of Nigeria (AUN).
The historical data comprised 978 undergraduate students from the AUN. The students
were termed as active students, i.e. students enrolled in the current semester. Table 1 Applying
shows that there is a decrease in students as the level of study increases which might be machine
associated with number of admissions and/or dropouts. An overwhelming number of the learning
students (91.7%) are between 16 and 20 years of age and the rest are above 20 years.
Pertaining to gender, 51.9% of the students are males and the rest are females. AUN is
approach
located in the North Eastern region of Nigeria, majority of the students (295, 39.5%) are
from this region followed closely by the North central region (150, 20.1%), which is the
closest region to the northeast in Nigeria. Consequently, 34.4% of the students are from
the Southern regions of Nigeria with the South-south region contributing to almost half
(115, 15.4%) of the students (Figure 1).
A total of 13 attributes were identified in the data set, namely, home address, telephone
number, email address, gender, admission date, date of birth, state of origin, country of
origin, JAMB score, level of study, marital status, name of previous school and CGPA. Six
irrelevant attributes for student performance prediction were deleted, namely, home
address, country of origin, telephone number, email address, marital status and name of
previous school. Country of origin was removed as 99.99% of the students were from
Nigeria and marital status was removed as only 0.02% of the students were married. In
total, 28 records were discarded due to erroneous entry and/or missing value. First year
students’ (n = 202) information were excluded due to lack of CGPA score, as they have not
earned any credit hours. Thus, seven relevant attributes were retained for analysis following
the recommendation of past studies (Ahmad et al., 2015; Garg, 2018; Hamoud et al., 2018;

Frequency (%)

Year
1 228 30.5
2 207 27.7
3 178 23.8
4 134 17.9
Entry age
15 20 2.7
16 157 21.0
17 262 35.1
18 164 22.0
19 74 9.9
20 28 3.7
21 24 3.2
22 9 1.2
23 3 0.4
26 2 0.3
27 2 0.3
28 1 0.1
29 1 0.1
Gender
Male 388 51.9
Female 359 48.1 Table 1.
CGPA Demographic
Fail 128 17.1 breakdown of the
Pass 619 82.9 students
K

Figure 1.
Student count by
geopolitical zones

Mohamed and Waguih, 2017; Putpuek et al., 2018; Saa et al., 2019). Preprocessing of the data
involved the following steps for the listed variables:
 Date of birth and admission date: These two attributes were used to obtain the
university entry age of the students and were subsequently standardized.
 Gender: The gender attribute was converted into male and female dummy variables
where 1 indicated the presence and 0 the absence of the male and female variables.
 State of origin: There are six geo-political zones in Nigeria which are divided
according to the ethnical preferences within the country. The Hausa ethnic groups
make up the north-west (NW) and north-east (NE), while the Yoruba ethnic groups
make up the south-west (SW), and south-east (SE) is made up of the Igbo ethnic
groups. The north-central (NC) and the south-south (SS) consist of ethnic minorities
in neighboring states with similar cultures. As every state can only be part of one
geopolitical zone, six new attributes (regions) were created (NW, NE, NC, SS, SW
and SE) and the values of 1 or 0 were assigned to indicate if a student was from the
geopolitical region or not respectively, i.e. dummy variables.
 JAMB score: JAMB score values are whole numbers between 0 and 400. Similar to
the entry age attribute, the JAMB scores were standardized.
 Level of study: Majority of the programs in the university are four-year programs
with the exception of the engineering and law programs which are usually five
years. As these programs are new to the university, the oldest students are in their
fourth year. Similar to the regions attribute, the values of 1 or 0 were assigned to
indicate if a student was in a particular year or not respectively (dummy variables).
 CGPA (Target variable): The CGPA, as mentioned earlier is on a scale of 4, thus a
CGPA value could be a number between 0 and 4 and rounded to 2 significant
figures. As a passing score is a CGPA of over 2.0, the value 1 was assigned to any
CGPA over 2.0 and 0 was assigned to values less than 2.0. Therefore, the target
variable, for the purpose of this study, is a discrete variable (1, 0), which is a
mandatory requirement for logistic regression methods.
3.2 Logistic regression Applying
The data preparation and processing resulted in six main attributes and 747 observations, machine
which are further processed into 14 attributes, namely, Entry Age, NC, NE, NW, SE, SS, SW,
Level 1, Level 2, Level 3, Level 4, Gender, JAMB and CGPA. The purpose of this study is to
learning
create a machine-learning algorithm that will predict if a student will achieve a passing approach
CGPA, i.e. over 2.0 on a 4.0 scale. Logistics regression, being a predictive analysis model, is
the preferred statistical method for this study as the dependent variable (CGPA) is
dichotomous. After preprocessing the data, 13 attributes were identified as the input
variables (Entry Age, NC, NE, NW, SE, SS, SW, Level 1, Level 2, Level 3, Level 4, Gender
and JAMB score) and one attribute as the target variable (CGPA). Scikit-learn (Sklearn)an
open-source machine-learning library with various algorithms such as clustering, regression
and classification was used in this study.
After loading the data set, the data set was randomly split in to 2: training data with 522
observations (70 %) and test data with 225 observations (30 %). From the 747 observations,
619 students had passing CGPAs while the remaining 128 had CGPAs below 2.0. Sklearn
was used to build and fit the logistic regression model using the training data. The observed
accuracy of the model based on the training data was 84.7% (0.8467432950191571). This
means that the model was 84.7% accurate in matching the outputs and the targets. Next, the
intercept, the coefficients of the variables, as well as their corresponding odds ratio
(exponential of the coefficients) were obtained as shown in Table 2.
From Table 2, the entry age variable seems to be the only variable that does not really
contribute to the model, as its odd ratio is approximately 1.0. Sklearn’s predict probability
(predict_proba) method was used to extract the probability of earning a CGPA (>2.0) for
each observation in the test data. Then a comma-delimited file (CSV) of the test cases was
extracted for a visual analysis of inputs versus their probabilities. Tableau software
(Tableau Desktop Professional 2020.1.2 version) was used to analyze the observations in the
test dataset along with their corresponding predictions obtained from running the model
with the test data set as mentioned above. Tableau is a sophisticated and interactive data
visualization application used to easily transform raw data into comprehensible formats.
Figures 2–6 show the output of the analysis between the variables and the predicted targets
performed using Tableau.

Input variables Coefficient Odds ratio

Year 4 1.713867 5.550385

Year 3 1.285781 3.617493
South West 0.952796 2.592948
JAMB Score 0.935355 2.548117
North Central 0.334934 1.397848
North West 0.292712 1.340057
Female 0.21119 1.235147
Entry Age 0.06319 1.065229
Male 0.211244 0.809576
North East 0.423244 0.654919
South East 0.543134 0.580925
South South 0.614118 0.541118 Table 2.
Year 2 1.243248 0.288446 Bias, variables
Year 1 1.756455 0.172656 coefficient and odds
Intercept (Bias) 2.983912 19.764993 ratio
K

Figure 2.
Probability of
academic success
versus entry age

Figure 3.
Probability of
academic success
versus gender

4. Results and discussion

This study sets out to use enrollment data to identify significant factors that can be used to
predict if students in a private university in Nigeria will achieve a passing CGPA. A
supervised machine-learning approach was adopted, and the data set was divided into
training data (70%) and a test data set (30%). Running a logistics regression model on the
training data, the model was trained and able to predict the students’ academic success/
 Applying
machine
learning
approach

Figure 4.
Probability of
academic success
versus JAMB score

Figure 5.
Student count by
geopolitical zones
(test data)

failure with an accuracy of 84.7%. The model was then used to the test data set and the
accuracy was observed to be 83.5%. Table 2 shows the bias (intercept) as well as the
coefficients of the input variables obtained from the trained model. The probability of
earning a passing CGPA for each observation in the test data set was extracted along with
the corresponding observations and analyzed with the Tableau visualization software. The
subsections below will discuss the analysis of the input variables and their associated
probabilities based on the test data set.
K

Figure 6.
Probability of
academic success
versus regions

4.1 Entry age

In this study, entry age signifies the age the student was accepted into university. The
influence of the variable entry age in the logistic regression model is negligible with an odds
ratio of 1.06 ( b = 0.06). Previous studies have examined the impact of age on performance of
students. Öner et al. (2018) observed that younger ages in children resulted in lower grades
in mathematics as well as inattention and higher hyperactivity. While Öner’s study was
based on children at a lower education level, the principle applies to university students
where academic performance has been observed to depend on demographic data such as age
(Fernandes et al., 2019; Helal et al., 2019). Figure 2 plots the average probability of academic
success versus the age of students based on the test data set. The entry ages of the students
captured in the test data set was between 15 and 26 years of age.
The graph in Figure 2 shows that for the age range of 15 to 21, the probability of
achieving a passing CGPA is between 80 and 90%. At the age of 22 years however, there is a
noticeable drop. The graph also shows that students above the age of 23 have the highest
chances of academic success; however, this could be misleading considering that there are
only two students in the test sample over the age of 23. Agreeing with previous research
(Mills et al., 2009; Amuda et al., 2016), the findings of this study suggest that age is not a
predictor of university students’ academic success. A possibility for this observation could
be that the entry age for majority (96.88%) of the students fall within the same age group (i.e.
youth between 15 and 24 years of age), Table 3, which has been defined as the active and
receptive age (Adelakun, 2017). Therefore, it is expected that the students in this age group
will have similar social and behavioral traits.
This finding supports the fact that as the level of study for the student increases, the
probability of success also increases. This is as a result of maturity that is associated with
the increase in age as well as the motivation to complete their studies as they approach
graduation.
4.2 Gender: male and female Applying
Table 4 shows that the gender of the students in the test data set is almost balanced, 113 are machine
male (50.2%), while there are 112 female students (49.80%). With an odds ratio of 1.23 ( b =
0.21), the female gender attribute is a significant variable in the logistic regression model.
learning
The male gender also contributes to the logistic regression model with an odds ratio of 0.81 approach
( b = 0.21). These findings indicate that the odds of academic success are 1.2 times more
likely if the student is a female. This supports previous research that shows that gender is a
predictor of academic success (Anderton and Chivers, 2016).
Figure 3 also shows that female students have a higher probability of achieving a
passing CGPA in comparison to male students. This finding is supported by previous
research (Reddy et al., 2017; Adelakun, 2017). Fischer et al. (2013) attributes this advantage
over male students to females showing a higher accomplishment motivation, stating that
female students show “more compensatory effort as well as self-control and taking more
pride in their own productivity which helps the female students to outperform their male
counterparts” (Fischer et al., 2013). Similarly, Spinath et al. (2014) believe that female
students are better suited for educational environments as they are verbally more intelligent,
stronger in terms of self-discipline and have a higher motivation (Spinath et al., 2014).

4.3 JAMB score

In the test data set, 208 out of the 226 students scored over the 180 JAMB entry score
requirement and in comparison, 213 students achieved a passing CGPA. With an odds ratio
of 2.55 ( b = 0.94), the JAMB score is an important feature in predicting academic success,
especially considering that the JAMB score variable was standardized. Admission into
Nigerian universities typically depends on cognitive entry characteristics of the students
that are mostly academic in nature such as the JAMB and the Senior School Certificate
Examination (SSCE) score. Adekitan and Noma-Osaghae (2019) suggest that cognitive entry
characteristics of students may not explain academic success, though in their data mining
study to predict the performance of first year university students, they observed that

Age Frequency (%)

15 7 3.11
16 47 20.89
17 81 36.00
18 44 19.56
19 29 12.89
20 9 4.00
21 3 1.33
22 3 1.33
23 1 0.44 Table 3.
26 1 0.44 Age distribution

Gender
Male 113 50.20 Table 4.
Female 112 49.80 Gender distribution
K students with a higher JAMB score performed better (Adekitan and Noma-Osaghae, 2019).
Bamgboye et al. (2001) also observed that students performed better in preclinical MBBS
examination than their counterparts with lower JAMB scores though their findings indicate
that WAEC/SSCE scores are a better predictor of the students’ performance. The trend
shown in Figure 4 suggests that the students’ probability of academic success increases
with an increasing JAMB score and corroborates the findings of Bamgboye et al. (2001) and
Adekitan and Noma-Osaghae (2019).

4.4 Region (ethnicity)

Nigeria has six geo-political zones based on ethnicity. Figure 5 shows the demographic
representation, with regards to the region/zones, of the students in the test dataset. Majority
of the students are from the northern part of Nigeria, specifically the northeast and north
central due to proximity to the university, which is in the northeast region. Our findings
show that the region is a significant predictor for the outcome variable. Students from the
southwest were having the highest probability of achieving a passing CGPA with a
coefficient of 0.95, followed by students from the north west ( b = 0.33) and the north central
( b = 0.30) regions.
The northeast, southeast and south-south regions had negative coefficients with odds
ratios of 0.65, 0.58 and 0.54 respectively. This indicates that for students in either of the three
regions, the odds of achieving a CGPA of 2.0 or more are 35, 43 and 46% lower than the base
model (Figure 6). Contrariwise to our findings, past research found that ethnic background
is not a predictor of students’ final graduation results (Adekitan and Salau, 2020). Their
study uses the following predictive algorithms: classification tree, random forest, Naïve
Bayes and neural networks. Similarly, Ojetunde et al. (2019) did not find evidence to support
that the geopolitical/ethnic background of students determines the degree outcome of
undergraduate students using a logistic regression model.

4.5 Level of study

In this study, the duration of all the programs available is 4 years with the exception of the
law and engineering programs, which last for 5 years. These two programs are relatively
new, and the most senior classes are year 4 and year 1, respectively. Table 5 shows the
number of students in each year within the test dataset. As expected, the frequency
decreases as the level increases due to students dropping out or not meeting the
requirements to enter the next level. Eddy (2016), suggests that as students’ progress in their
academics, the courses they take tend to be less general and more difficult in nature. This
would be one of the reasons why there is a drop in the number of students as their education
level increases.
The Year 4 ( b = 1.71) and Year 3 ( b = 1.21) variables have the largest coefficients in the
logistic regression model. This is an expected result as the students at this stage will be
more dedicated and put in more effort not only to graduate but also to improve their CGPA

Frequency (%)

Year
Year 1 73 32.40
Table 5. Year 2 60 26.70
Study year Year 3 52 23.10
distribution Year 4 40 17.80
to obtain a higher classed degree. As expected, the Year 1 variable ( b = 1.24) has the Applying
lowest coefficient followed by Year 2 ( b = 1.76) among the student years as this is where machine
students are adjusting to the demands of a new academic and social environment (Junco,
2015). While it is expected that the courses students take become more difficult as they
learning
progress in their education, the probability of them achieving a passing CGPA increases as approach
they are more determined to excel and those who are less likely to succeed would have
dropped out of the program (Figure 7).

4.6 Implications for theory and practice

In this study, a supervised machine-learning approach was used to predict students’
academic success using enrolment data. The enrolment data was trained to accurately
predict 84.7% of the students CGPA using the logistics regression algorithm. The algorithm
was applied to a test data resulting in 83.5% accuracy. The most significant contributors in
the model, of decreasing magnitude, include the following variables: Year 4, Year 3, SW,
JAMB Score, Gender, NC and NW. The present work unveils catchy theoretical implications.
A burgeoning stream of research highlighted several limitations of self-reported survey data
and LMS data in predicting students’ academic success (Conijn et al., 2016; Rincon-Flores
et al., 2020; Tempelaar et al., 2020; Tempelaar et al., 2015).
In light of these reasonings, this paper strives to predict students’ academic success
using machine learning approach and enrolment data from SIS. In doing so, this paper
responds to research calls and adds insights to this line of research. Theoretically, a
complementary perspective is needed to move beyond the approach of emphasis on one or
two predictors of student academic performance toward complementary considerations.
Thus, this paper adds to the literature by showing the value and complementary nature of
several predictors in Sub-Saharan region, where limited empirical findings are available.
This work also paves the way for future work towards exploration of potential moderators

Figure 7.
Probability of
academic success
versus level of study
K and mediators. We envisage research opportunities that leverage other machine learning
algorithms.
This present study divulged several complementary practical implications. Structural
disadvantage appears to have adverse effects on student outcomes. Past work in the UK
echoed similar arguments, as BAME (Black, Asian and Minority Ethnic) students from
lower socio-economic backgrounds were less likely to attain academic success (Crawford,
2014; Soria and Stebleton, 2012). The results show that students from the northeast,
southeast and south-south do not perform as well as students from the northwest, south
west and north central. This could be attributed to the quality of the universities and
teachers, which have been affected by conflicts, such as Boko Haram in the North east
(Bertoni et al., 2019; Abiodun et al., 2020) and Niger Delta militants in the south-south region
(Pepple and Ogologo, 2017).
Technically, the three regions can be characterized as having structural disadvantage,
that are inflated by other factors such as community and engagement (influence student’s
intellect, behavior and desire for studies) and finances in the south east. Similar arguments
were echoed by (Crawford, 2014; Morrow and Ackermann, 2012). Federal, state and local
governments need to do more to support students in these regions. One, interventions
programs to develop the quality of tutors, tutoring and learning activities should be
implemented in the region. Two, government needs to increase educational investments and
staff posting, for instance, experience and high performing professors and lecturers should
be posted to these regions to revive the educational system in the region starting from
primary to tertiary level. Alternatively, pay increase can be used in the affected regions to
attract teachers, lecturers and professors to work in the regions.
Three, the ability to identify students who are at risk of achieving a passing CGPA is
important for educators as they can apply timely intervention measures to aid the students
which can lead to a reduction of attrition rates. For instance, based on the findings of this
study, universities are advised to focus on students at risk, e.g. Years 1 and 2 students. The
academic advising unit as well as the school psychologist can arrange sessions with the
students that aims at helping them adjust to the demands of a new academic and social
environment. Societies, social and cultural students’ clubs are facilitators for greater
integration among new students, as integration increases, information flows within the
students which could serve as a helper in academic activities. At faculty and departmental
level, peer-to-peer tutoring can be offered where high performing students can assist new,
struggling and/or students at risk of failing.
Four, apart from encouraging peer-to-peer tutoring, implementing this model could help
universities manage intellectual talent, in that academically gifted and motivated students
are to be identified, attracted, groomed and guided to pursue their career of interests, who
might be attracted to other universities. In particular, their accomplishments can not only
create an academically vigorous climate for competition and knowledge sharing but also
boost the universities’ prestige and reputation as highlighted in past work (Miguéis et al.,
2018).
Fifth, the results from this study can also help in recruitment of students. The ability to
identify students who are at risk of achieving a passing CGPA is important for educators as
they can apply timely intervention measures to aid the students, which can lead to a
reduction of attrition rates. Most universities used to accept a minimum JAMB score of 180,
this study shows that majority of the students that scored between 160 and 180 falls into the
group that has more than a 50% chance of achieving a passing CGPA. Therefore, admission
teams can use the findings of this study to improve their decision-making process for
admitting students into the university. Past works made similar suggestions about the
opportunities surrounding knowledge management and organizational intelligence (Danõa Applying
et al., 2020; Holley, 2009; Metcalfe, 2005). machine
learning
4.7 Strengths, limitations and future research direction
Applicable to most research, this study has several limitations and strengths that should be
approach
noted. In terms of limitations, limited number of variables were available in the student
information system at the time of the study. For instance, West African Examinations
Council (WAEC) results were not considered, and past work shows WAEC results as
important predictors (Bamgboye et al., 2001). Second, other important variables such as
personality traits, stress, anxiety and motivation are not captured which limits our ability to
draw sound conclusions. Future studies are encouraged to consider pre-enrollment data (i.e.
WAEC results) and other variables, for instance, a mixture of secondary that captures
demographic data and primary data that captures perceptions, can be used to model and
predict students’ academic performance.
Third, the present work uses one machine learning algorithm (i.e. logistics regression).
Future studies can compare the logistics regression model with other machine learning
models, specifically, classification methods such as neural networks, decision trees, random
forests, for additional insights. Past work showed that a combination of machine learning
techniques can help to compare the accuracy and performance and identify the suitable
techniques for any given situation (Rusli et al., 2008; Amirhajlou et al., 2019). In terms of
strengths, self-reported and cross-sectional data are vulnerable to common method variance
(Podsakoff et al., 2012). This study uses secondary and multisource data to evade the
potential threat of common method variance and to increase our ability to draw concrete
causal inference. Finally, the use of machine learning technique increases the reliability and
validity of our findings, which is almost impossible using conventional methods.

References
Abiodun, T.F., Omolayo, O.O., Tomisin, A.D. and Chinedu, O.C. (2020), “Assessment of boko haram
insurgents’ threats to educational development in the northeast Nigeria: the way forward”,
Assessment, Vol. 3 No. 1, pp. 31-43.
Adekitan, A.I. and Noma-Osaghae, E. (2019), “Data mining approach to predicting the performance of
first year student in a university using the admission requirements”, Education and Information
Technologies, Vol. 24 No. 2, pp. 1527-1543.
Adekitan, A.I. and Salau, O. (2020), “Toward an improved learning process: the relevance of ethnicity to
data mining prediction of students’ performance”, SN Applied Sciences, Vol. 2 No. 1, p. 8.
Adelakun, O.E. (2017), “Gender disparity in academic performance of students in the faculty of
agriculture and forestry, university of Ibadan, Oyo state”, International Journal of Agricultural
Economics and Rural Development, Vol. 9 No. 1.
Ahmad, F., Ismail, N.H. and Aziz, A.A. (2015), “The prediction of students’ academic performance using
classification data mining techniques”, Applied Mathematical Sciences, Vol. 9 No. 129,
pp. 6415-6426.
Akgül, A. (2018), “A novel method for a fractional derivative with non-local and non-singular kernel”,
Chaos, Solitons and Fractals, Vol. 114, pp. 478-482.
Akgül, E.K. (2019), “Solutions of the linear and nonlinear differential equations within the generalized
fractional derivatives”, Chaos: An Interdisciplinary Journal of Nonlinear Science, Vol. 29 No. 2,
p. 023108.
Al-Sudani, S. and Palaniappan, R. (2019), “Predicting students’ final degree classification using an
extended profile”, Education and Information Technologies, Vol. 24 No. 4, pp. 2357-2369.
K Alyahyan, E. and Düs tegör, D. (2020), “Predicting academic success in higher education: literature
review and best practices”, International Journal of Educational Technology in Higher Education,
Vol. 17 No. 1.
Amirhajlou, L., Sohrabi, Z., Alebouyeh, M.R., Tavakoli, N., Haghighi, R.Z., Hashemi, A. and Asoodeh,
A. (2019), “Application of data mining techniques for predicting residents’ performance on pre-
board examinations: a case study”, Journal of Education and Health Promotion, Vol. 8.
Amuda, B.G., Bulus, A.K. and Joseph, H.P. (2016), “Marital status and age as predictors of academic
performance of students of colleges of education in the North-Eastern Nigeria”, American
Journal of Educational Research, Vol. 4 No. 12, pp. 896-902.
Anderton, R. and Chivers, P. (2016), “Predicting academic success of health science students for first
year anatomy and physiology”, International Journal of Higher Education, Vol. 5 No. 1.
Aulck, L., Nambi, D., Velagapudi, N., Blumenstock, J. and West, J. (2019), “Mining university registrar
records to predict First-Year undergraduate attrition”, International Educational Data Mining
Society.
Baleanu, D., Fernandez, A. and Akgül, A. (2020), “On a fractional operator combining proportional and
classical differ integrals”, Mathematics, Vol. 8 No. 3, p. 360.
Bamgboye, E.A., Ogunnowo, B.E., Badru, O.B. and Adewoye, E.O. (2001), “Students admission grades
and their performance at Ibadan university pre-clinical MBBS examinations”, African Journal of
Medicine and Medical Sciences, Vol. 30 No. 3, pp. 207-211.
Bawack, R.E. and Kamdjoug, J.R.K. (2020), “The role of digital information use on student performance
and collaboration in marginal universities”, International Journal of Information Management,
Vol. 54, p. 102179.
Berens, J., Schneider, K., Görtz, S., Oster, S. and Burghoff, J. (2018), “Early detection of students at risk–
predicting student dropouts using administrative student data and machine learning methods”,
Journal of Educational Data Mining, Vol. 11 No. 3, pp. 1-41.
Bertoni, E., Di Maio, M., Molini, V. and Nistico, R. (2019), “Education is forbidden: the effect of the boko
haram conflict on education in North-East Nigeria”, Journal of Development Economics, Vol. 141,
p. 102249.
Bramer, M. (2016), Principles of Data Mining, Springer, London.
Conijn, R., Snijders, C., Kleingeld, A. and Matzat, U. (2016), “Predicting student performance from LMS
data: a comparison of 17 blended courses using moodle LMS”, IEEE Transactions on Learning
Technologies, Vol. 10 No. 1, pp. 17-29.
Cooper, H., Carlisle, C., Gibbs, T. and Watkins, C. (2001), “Developing an evidence base for
interdisciplinary learning: a systematic review”, Journal of Advanced Nursing, Vol. 35 No. 2,
pp. 228-237.
Crawford, C. (2014), “Socio-economic differences in university outcomes in the UK: drop-out, degree
completion and degree class (no. W14/31)”, IFS Working Papers.
Danõa, J., Caputo, F. and Racek, J. (2020), “Complex network analysis for knowledge management and
organizational intelligence”, Journal of the Knowledge Economy, Vol. 11 No. 2, pp. 405-424.
Eddy, T.M. (2016), Reflecting Back: Do Senior Students Believe They Experienced a Sophomore Slump?.
Fernandes, E., Holanda, M., Victorino, M., Borges, C., Carvalho, R. and Van Erven, G. (2019),
“Educational data mining: predictive analysis of academic performance of public-school
students in the capital of Brazil”, Journal of Business Research, Vol. 94, pp. 335-343.
Fischer, F., Schult, J. and Hell, B. (2013), “Sex differences in secondary school success: why female
students perform better”, European Journal of Psychology of Education, Vol. 28 No. 2,
pp. 529-543.
Francis, P., Broughan, C., Foster, C. and Wilson, C. (2020), “Thinking critically about learning analytics,
student outcomes, and equity of attainment”, Assessment and Evaluation in Higher Education,
Vol. 45 No. 6, pp. 811-821, doi: 10.1080/02602938.2019.1691975.
Garg, R. (2018), “Predict student performance in different regions of Punjab”, International Journal of Applying
Advanced Research in Computer Science, Vol. 9 No. 1, pp. 236-241.
machine
Gaševic, D., Dawson, S., Rogers, T. and Gasevic, D. (2016), “Learning analytics should not promote one
size fits all: the effects of instructional conditions in predicting academic success”, The Internet
learning
and Higher Education, Vol. 28, pp. 68-84. approach
Hamoud, A.K., Hashim, A.S. and Awadh, W.A. (2018), “Predicting student performance in higher
education institutions using decision tree analysis”, International Journal of Interactive
Multimedia and Artificial Intelligence, Vol. 5 No. 2.
Helal, S., Li, J., Liu, L., Ebrahimie, E., Dawson, S. and Murray, D.J. (2019), “Identifying key factors of
student academic performance by subgroup discovery”, International Journal of Data Science
and Analytics, Vol. 7 No. 3, pp. 227-245.
Hoffait, A.S. and Schyns, M. (2017), “Early detection of university students with potential difficulties”,
Decision Support Systems, Vol. 101, pp. 1-11.
Holley, K.A. (2009), “Understanding interdisciplinary challenges and opportunities in higher
education”, ASHE Higher Education Report, Vol. 35 No. 2, pp. 1-131.
Huang, A.Y., Lu, O.H., Huang, J.C., Yin, C.J. and Yang, S.J. (2020), “Predicting students’ academic
performance by using educational big data and learning analytics: evaluation of classification
methods and learning logs”, Interactive Learning Environments, Vol. 28 No. 2, pp. 206-230.
Junco, R. (2015), “Student class standing, Facebook use, and academic performance”, Journal of Applied
Developmental Psychology, Vol. 36, pp. 18-29.
Kamal, P. and Ahuja, S. (2019), “Academic performance prediction using data mining techniques:
identification of influential factors effecting the academic performance in undergrad
professional course”, Harmony Search and Nature Inspired Optimization Algorithms, Springer,
Singapore, pp. 835-843.
Kanyip, B.P. (2013), Admission Crisis in Nigerian Universities: The Challenges Youth and Parents Face
in Seeking Admission.
Kovacic, Z. (2010), Early Prediction of Student Success: Mining Students’ Enrolment Data.
Kumari, K. and Yadav, S. (2018), “Linear regression analysis study”, Journal of the Practice of
Cardiovascular Sciences, Vol. 4 No. 1, p. 33.
Lee, T.R., Wood, W.T. and Phrampus, B.J. (2019), “A machine learning (kNN) approach to predicting
global seafloor total organic carbon”, Global Biogeochemical Cycles, Vol. 33 No. 1, pp. 37-46.
Metcalfe, A. (Ed.), (2005), Knowledge Management and Higher Education: A Critical Analysis: A Critical
Analysis, IGI Global.
Miguéis, V.L., Freitas, A., Garcia, P.J. and Silva, A. (2018), “Early segmentation of students according to
their academic performance: a predictive modelling approach”, Decision Support Systems,
Vol. 115, pp. 36-51.
Mills, C., Heyworth, J., Rosenwax, L., Carr, S. and Rosenberg, M. (2009), “Factors associated with the
academic success of first year health science students”, Advances in Health Sciences Education,
Vol. 14 No. 2, pp. 205-217.
Mohamed, M.H. and Waguih, H.M. (2017), “Early prediction of student success using a data
mining classification technique”, International Journal of Science and Research, Vol. 610,
pp. 126-131.
Moreno-Marcos, P.M., Pong, T.C., Muñoz-Merino, P.J. and Kloos, C.D. (2020), “Analysis of the factors
influencing learners’ performance prediction with learning analytics”, IEEE Access, Vol. 8,
pp. 5264-5282.
Morrow, J. and Ackermann, M. (2012), “Intention to persist and retention of first-year students: the
importance of motivation and sense of belonging”, College Student Journal, Vol. 46 No. 3,
pp. 483-491.
K Musumeci, F., Rottondi, C., Nag, A., Macaluso, I., Zibar, D., Ruffini, M. and Tornatore, M. (2018), “An
overview on application of machine learning techniques in optical networks”, IEEE
Communications Surveys and Tutorials, Vol. 21 No. 2, pp. 1383-1408.
Mwalumbwe, I. and Mtebe, J.S. (2017), “Using learning analytics to predict students’ performance in
moodle learning management system: a case of Mbeya university of science and technology”,
The Electronic Journal of Information Systems in Developing Countries, Vol. 79 No. 1, pp. 1-13.
Ojetunde, I., Sule, A.I., Kemiki, O.A. and Olatunji, I.A. (2019), “Factors affecting the academic
performance of real estate students in a specialized federal university of technology in Nigeria”,
Property Management, Vol. 38 No. 2.
Olaya, D., Vasquez, J., Maldonado, S., Miranda, J. and Verbeke, W. (2020), “Uplift modeling for
preventing student dropout in higher education”, Decision Support Systems, Vol. 134, p. 113320.
ır, Ö., Ayyıldız, N., Çelika
Öner, Ö., Çalıs g, I., Uran, P., Olkun, S. and Çiçek, M. (2018), “Effects of changed
school entry rules: age effects within third grade students”, EURASIA Journal of Mathematics,
Science and Technology Education, Vol. 14 No. 6, pp. 2555-2562.
Oyedeji, A.O., Salami, A.M., Folorunsho, O. and Abolade, O.R. (2020), “Analysis and prediction of
student academic performance using machine learning”, JITCE (Journal of Information
Technology and Computer Engineering), Vol. 4 No. 1, pp. 10-15.
Paliwal, M. and Kumar, U.A. (2009), “A study of academic performance of business school graduates
using neural network and statistical techniques”, Expert Systems with Applications, Vol. 36
No. 4, pp. 7865-7872.
Pepple, T.F. and Ogologo, G.A. (2017), “The effects of the Niger Delta crisis on educational resources,
attitude to schooling, and academic achievement of basic science students in Rivers state,
Nigeria”, Journal of the International Society for Teacher Education, Vol. 21 No. 1, pp. 67-76.
Podsakoff, P.M., MacKenzie, S.B. and Podsakoff, N.P. (2012), “Sources of method bias in social science
research and recommendations on how to control it”, Annual Review of Psychology, Vol. 63 No. 1,
pp. 539-569.
Putpuek, N., Rojanaprasert, N., Atchariyachanvanich, K. and Thamrongthanyawong, T. (2018),
“Comparative study of prediction models for final GPA score: a case study of rajabhat
rajanagarindra university”, 2018 IEEE/ACIS 17th International Conference on Computer and
Information Science, pp. 92-97.
Reddy, B.V., Gupta, A. and Singh, A.K. (2017), “A study to assess factors affecting the performance of
undergraduate medical students in academic examination in community medicine”, Int J Med.
Public Health, Vol. 4 No. 4, pp. 1066-1070.
Rincon-Flores, E.G., Lopez-Camacho, E., Mena, J. and Lopez, O.O. (2020), April). “Predicting academic
performance with artificial intelligence (AI), a new tool for teachers and students”, 2020 IEEE
Global Engineering Education Conference (EDUCON), IEEE, pp. 1049-1054.
Rizvi, S., Rienties, B. and Khoja, S.A. (2019), “The role of demographics in online learning; a decision
tree based approach”, Computers and Education, Vol. 137, pp. 32-47.
Rusli, N.M., Ibrahim, Z. and Janor, R.M. (2008), “Predicting students’ academic achievement:
comparison between logistic regression, artificial neural network, and neuro-fuzzy”, 2008
International Symposium on Information Technology, IEEE, Vol. 1, pp. 1-6.
Saa, A.A., Al-Emran, M. and Shaalan, K. (2019), “Mining student information system records to predict
students’ academic performance”, International Conference on Advanced Machine Learning
Technologies and Applications, Springer, Cham, pp. 229-239.
Sedrakyan, G., Malmberg, J., Verbert, K., Järvelä, S. and Kirschner, P.A. (2020), “Linking learning
behavior analytics and learning science concepts: designing a learning analytics dashboard for
feedback to support learning regulation”, Computers in Human Behavior, Vol. 107, p. 105512.
Shahiri, A.M. and Wahidah, H. (2015), “A review on predicting student’s performance using data
mining techniques”, Procedia Computer Science, Vol. 72, pp. 414-422.
Shum, S.B. and Ferguson, R. (2012), “Social learning analytics”, Journal of Educational Technology Applying
&Society, Vol. 15 No. 3, pp. 3-26.
machine
Sin, K. and Muthu, L. (2015), “Application of big data in education data minning and learning
analytics–a literature review”, ICTACT Journal on Soft Computing, Vol. 5 No. 4. learning
Soofi, A.A. and Awan, A. (2017), “Classification techniques in machine learning: applications and approach
issues”, Journal of Basic and Applied Sciences, Vol. 13, pp. 459-465.
Soria, K.M. and Stebleton, M.J. (2012), “First-generation students’ academic engagement and retention”,
Teaching in Higher Education, Vol. 17 No. 6, pp. 673-685.
Spinath, B., Eckert, C. and Steinmayr, R. (2014), “Gender differences in school success: what are the
roles of students’ intelligence, personality and motivation?”, Educational Research, Vol. 56 No. 2,
pp. 230-243.
Tempelaar, D.T., Rienties, B. and Giesbers, B. (2015), “In search for the most informative data for
feedback generation: Learning analytics in a data-rich context”, Computers in Human Behavior,
Vol. 47, pp. 157-167.
Tempelaar, D., Rienties, B. and Nguyen, Q. (2020), “Subjective data, objective data and the role of bias in
predictive modelling: lessons from a dispositional learning analytics application”, Plos One,
Vol. 15 No. 6, p. e0233977.
Yaacob, W.F., Nasir, S.A., Yaacob, W.F. and Sobri, N.M. (2019), “Supervised data mining approach for
predicting student performance”, Indonesian Journal of Electrical Engineering and Computer
Science, Vol. 16 No. 3, pp. 1584-1592.

Further reading
Akçapınar, G., Altun, A. and As kar, P. (2019), “Using learning analytics to develop early-warning
system for at-risk students”, International Journal of Educational Technology in Higher
Education, Vol. 16 No. 1, p. 40.

Corresponding author
Mohammed Nasiru Yakubu can be contacted at: [email protected]

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: [email protected]

View publication stats