Case Study International Journal of Engineering Technologies and Management Research

ISSN (Online): 2454-1907 February 2022 9(2), 64-81

PREDICTIVE ANALYTICS OF ACADEMIC PERFORMANCE OF SENIOR

HIGH SCHOOL (SHS) STUDENTS: A CASE STUDY OF SUNYANI SHS
Adjei-Pokuaa Henrietta 1 , Adebayo F. Adekoya 2

1, 2 Department of Computer Science and Informatics, University of Energy and Natural Resources, Ghana

ABSTRACT
Due to the availability and increasing adoption of technology in learning
management systems, online admission systems, school management systems, and
educational databases have expanded in recent years.
Motivation/Background: Literature shows that these data contain vital and relevant
information that could be used to monitor and advise students’ so that their
performance could be enhanced. In this study, the random forest algorithm is
proposed to identify and examine the factors that influence students’ performance in
WASSCE. Also, predict the future performance of students in WASSCE.
Method: A total of one thousand five hundred and twenty students’ data were
selected from Sunyani SHS. The results revealed that demographic data (age and
gender) do not influence the performance of students in their final WASSCE.
Results: However, an accuracy of 89.4% with error metrics (RMSE) 0.001639 and
Received 03 January 2022 MAPE error of 0.001321 revealed that the proposed model could effectively predict
Accepted 01 February 2022
the performance of students in the WASSCE.
Published 21 February 2022
Corresponding Author Keywords: Learning Management, Demographic Data, Effectively Predict, Random
Adjei-Pokuaa Henrietta, Forest Algorithm
herttydonkor@[Link]
DOI 10.29121/ijetmr.v9.i2.2022.1088
Funding: This research received no 1. INTRODUCTION
specific grant from any funding agency in
the public, commercial, or not-for-profit The size of the educational database keeps proliferating every three
sectors. years and these databases contain useful information that can be effectively
Copyright: © 2022 The Author(s). This is employed to improve the academic performance of students Yadav and Pal
an open access article distributed under (2012). There is a growing focus on institutional data mining by
the terms of the Creative Commons administrators, educational planners, and managers due to the exponential
Attribution License, which permits growth of educational data. (Anuradha & Velmurugan, 2015). Data mining
unrestricted use, distribution, and
reproduction in any medium, provided
techniques have been applied in educational systems to increase the
the original author and source are understanding of the process of learning by concentrating on key issues such
credited. as identifying, mining, and evaluating parameters and variables which are
associated with the learning progress of students Yadav and Pal (2012).
Students’ academic performance is a critical factor in every educational
setting, especially in senior high learning institutes, because educational
institutions are rated based on the academic excellence achieved by their
students and academic staff Mohamed et al. (2016). Forecasting the academic
performance of students in the real world has many challenges, but this has
been made easy with the growth in information and communication
technologies granting access to a large amount of information that could
facilitate the critical decision-making process. Knowledge Discovery in
How to cite this article (APA): Henrietta, A. P., and Adekoya, A. F. (2022). Predictive Analytics of Academic Performance of Senior 64
High School (Shs) Students: A Case Study of Sunyani Shs. International Journal of Engineering Technologies and Management
Research, 9(2), 64-81. doi: 10.29121/ijetmr.v9.i2.2022.1088
Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

Databases (KDD) also called Data Mining (DM) is the process of extracting hidden
patterns and discovering of associations between parameters in a massive amount
of data (Ahmad et al., 2015; Kaur et al., 2015). DM techniques have made numerous
achievements in areas such as marketing, education, engineering, finance,
healthcare, and sports by helping decision-makers in finding a solution to everyday
problems in these fields. This study focused on predicting final WASSCE score of SHS
students’ using random forest algorithm based on their Demographic Data and Past
Exams records. Undoubtedly, this investigation will lead to students’ performance
prediction based on their demographic data and previous final examinations score,
and this will ensure an immense benefit as follows: By knowing this information,
teachers will be able to take the needed actions of the learning activities in the
classroom to enhance students' skills. Enhance better arrangement of sitting
position in class to maintain a balanced education and promote students to student
interaction among fast and slow learners. Knowing the performance of students at
the early stage helps teachers and schools’ authorities to take charge and provide
the needed assistance to the students involved
General Factors Affecting the Academic Performance of Students: Literature
has shown that academic is affected by four main factors and that includes: the effect
of teaching and learning materials on students’ academic performance, the effect of
administrative power of organizational practices on students’ academic
performance, the consequence of teacher related factors on students’ academic
performance, and finally the effect of socioeconomic background on students’
academic performance. Teaching and learning resources, and environment have
great impact on students’ Academic Performance. According to Kieti (2017)
educational creativity of teachers and students is believed to be a dependent of
available facilities within the educational settlement or environments. Abundance
in teaching and learning facilities and resources such as physical facilities, which
include the habitable state-of-class and lecture-rooms, computer-room, equipped-
library, laboratories, better dining halls, qualified and experienced teachers,
teaching and learning materials (TLMs) contributes immensely to the success of
students. As stated by Kieti (2017), Tran et al. (2017), Tuen et al. (2019) sufficiency
current textbooks, learning videos and software, and other materials coincidental to
learning have a good reflection on students’ performance when these studying
materials are supplied or provided sufficiently. As proposed by (Bydovska &
Popelinsky, 2013, Li et al. (2013) concluded that an excellent illumination, clean and
fresh air, safe, quiet, and comfortable learning environs are essential factors for
positive achievement academically by students. It has been confirmed by (James et
al., 2015, Morrison, 2017, Tuen et al. (2019) that, encouraging physical exercises,
diminishing stress, decreasing air and noise pollution, and exposure to heat is
excellent teaching and learning atmosphere. Studies have presented those
organizational practices have effects on students’ academic performance such
practice include but not limited to instructions, leadership, and management Kieti
(2017), Buabeng-Andoh, 2015, Egbenya and Halm (2016) Academic scholars have
presented in literature that teacher-related factors such as the teacher’s
commitment, motivation, workload, and frequency of absenteeism (Affum-osei et
al., 2014) have effects on students’ academic performance.
On the other hand, socio-economic background of the student: the Socio-
Economic Status (SES) of a child is generally determined by merging parents’
occupational status, educational level, and income level. Research has confirmed
over and over that, SES affects students’ academic performance and that students’
who have a low SES earn lower exams and quiz scores and are more likely to be
International Journal of Engineering Technologies and Management Research 65
Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

school dropout (Al-Rahmi & Zeki, 2017; Oladejo et al., 2011; Seah et al., 2015). The
consequence of SES on students’ academic performance has been found to dominate
other educational influential factors, like parental involvement. There are highly
negative associations with low SES and academic performance of students. Other
scholars have argued that, economically some parents are less capable of paying for
the cost of education of their children at advanced levels and subsequently, their
children do not work at their maximum potentials (Nguyen, 2017). The home
environs have a remarkably significant part to play on the academic performance of
every child.
Application of Data Mining in Academic Performance Prediction: Knowledge of
students’ performance in advance is key in the educational settings Bhardwaj and
Pal (2011), Kaur et al., 2015; Ming et al., 2014). The performance of a student is
pivoted on many dynamics like personal, psychological, social, and other
environmental features. An effective and efficient way to achieve students’ academic
performance prediction is the use of the technique of Data Mining (DM) mostly
referred to as Knowledge Discovery in Databases (KDD) (Ahmad et al., 2015; Kaur
et al., 2015). DM techniques are applied in an extensive database to notice unseen
relationships and patterns helpful in decision making (Argiddi & Apte, 2012; Kaur
et al., 2015; Khasanah and Harwati (2017), Ramesh et al., 2013). DM can also be
employed in educational institutions to promote understanding of the pedagogical
and learning process to focus on discovery and evaluating the variables related to
students learning Bhardwaj and Pal (2011), Devasia et al. (2016). DM applies to
Artificial Intelligence (AI) and machine learning algorithms in discovering these
hidden details in the educational database.
Numerous researchers have applied machine learning (ML) algorithms for
predicting and detecting factors that influence student academic performance.
Literature shows that the academic success of students is dependent on one factor
or the other. Hence, there is no dependable agreement among different studies
Table 1 present details of input data for prediction students’ academic performance.
Here are some of these studies. The student performance prediction system
based on a decision support system (DSS) was proposed by (Dole & Rajurkar, 2014)
using the Naïve Bayes algorithm. A novel DSS framework for predicting students’
performance concerning their final examinations of a school was carried out in
(Livieris et al., 2015) using a hybrid machine learning algorithm. A hybrid classifier
consisting of three parallel algorithms, namely DT, KNN, and Aggregating One-
Dependence Estimators (AODE) in Pandey and Taruna (2016) was implemented for
accurate prediction of students’ performance. An assembling of three DT (C4-5, ID3,
and CART) algorithms were applied on the data of students studying engineering to
predict their performance in their final examinations in Yadav and Pal (2012).
Engineering students’ academic performance was carried out in Li et al. (2013)
by lump together machine learning algorithms namely principal component
analysis, K-means, and hierarchical clustering, and KNN and NB classifiers. The
intellectual and non-intellectual actions of students, together with demographic
information was used as input parameters to a predictive framework proposed by
Musso et al. (2013) to predict students’ academic performance based on Artificial
Neural Networks (ANN). A combination of NB, Multilayer Perceptron (MLP), and DT
algorithms was proposed in Osmanbegović et al. (2014), Osmanbegovic and Suljic
(2012) to predict students’ academic performance. A decision tree classifier was
implemented to predict the performance of students in Berhanu (2015).

International Journal of Engineering Technologies and Management Research 66

Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

Students’ performance prediction model using Multi-Agent Data Mining was

proposed in (AlMalaise et al., 2014) to predict the performance of the students
based on their past academic records. ANN model, for predicting the probable
performance of an applicant of university admissions was implemented and tested
Oladokun et al. (2008). Similarly, ANN-based student academic performance
predictive framework with two meta-heuristic algorithms motivated by cuckoo
birds and their lifestyle, namely, Cuckoo Search (CS) and Cuckoo Optimization
Algorithm (COA) is proposed in Chen et al. (2014).
A Neural Network model is proposed to predict the performance of students in
an academic organization Agrawal and Mavani (2015). A combination of ANN, DT,
and Support Vector Machines (SVMs) was assembled in Tran et al. (2017) to predict
students’ academic performance. The precise predictions of learners’ academic
performance at the early stages of the academic programme aids in the
identification of the weak students and assist management and teachers in taking
the counteractive measures to prevent them from failure. The performance of a
Senior High School (SHS) student is a matter of importance because this determines
whether the student will get a better grade to further his/her education in a higher
learning institution.

Table 1 Input Data for ML in Predicting Students’ Academic Performance

Reference Inputs variable Description

Kieti (2017), Musso et al. (2013) Age Demographic Data
Chen et al. (2014), Musso et al. (2013), Oladokun et al. Gender
(2008)
Devasia et al. (2016), Goga et al. (2015), Musso et al. Father’s
(2013), Yadav and Pal (2012), Yusif et al. (2011) occupation and
Mother’s
occupation
Adejo and Connolly (2018), Attuquayefio et al. (2014), Ethnicity
Tuen et al. (2019)
Cortez and Silva (2008), Goga et al. (2015) Parents’ marital
status
Agrawal and Mavani (2015), Bhardwaj and Pal (2011), Family size
Devasia et al. (2016), Goga et al. (2015), Khasanah and
Harwati (2017), Osmanbegović et al. (2014),
Osmanbegovic and Suljic (2012), Yadav and Pal (2012),
Yeboah (2014), Yusif et al. (2011)
Egbenya and Halm (2016), Fleischer (2015), Mohamed Exam’s score Past Exams
et al. (2016), Oladokun et al. (2008) records
Proposed Model Age, Gender, Demographic Data
Exam’s score, & Past Exams
Ethnicity records

2. MATERIALS AND METHODS

This study focused is to build an efficient predictive model for final WASSCE
score of SHS students’ using random forest algorithm based on their Demographic
Data and Past Exams records. The methods, procedures and techniques adopted are
organized into the following sections study design, area of the study, the population
of the study, sample and sampling technique, instruments for data collection, the
validity of the instruments, reliability of the instruments, method of data
collection/instrument, and method of data presentation/analysis.

International Journal of Engineering Technologies and Management Research 67

Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

3. PROPOSED CONCEPTUAL FRAMEWORK

The framework for predicting the academic performance of students’ using the
RF algorithm, as shown in Figure 1. The framework illustrates the steps involved in
developing models to predict final exams score for SHS final year students. There
are three main phases involved in this research which are data-collection, data-
integration, data-transformation, and pattern extraction. To apply DM algorithms
on any data, it is essential to carry out some data pre-processing tasks such as data-
cleaning, integration, discretization, and data-transformation Berhanu (2015).
Data Collection and Integration: 1520 participants from records of student’s
offering Business, General Art, Visual Art, Home Economics, and Science, which
consisted of both demographic and past academic performance were integrated into
one file using SQL database server at this stage as shown in Figure 1.
Data Transformation: At this stage, the integrated data was passed through
three distinct stages, (i) data cleaning which includes filling in missing values,
smoothing noise, identification, and removing of outliers where necessary and
resolving data inconsistency. (ii) Data normalization and aggregation (iii) data
reduction, where volumes of data were reduced but produce the same or similar
analytical results using feature selection and feature extraction. All data
transformation process was carried out with Python. Patterns Extraction: Weka, an
open-source data mining program, was used to construct the predictive model using
the RF regressor.
Data Partitioning: 1500 out of the 1520 clean data was partitioned into two,
thus training dataset (Train_D) which accounted for 75% of the dataset (DS) and the
remaining 25% was apportioned for testing (Test_D) the proposed model as shown
in figure 3.2. The reaming 20 records were the records of students who are still in
school (SHS 3) preparing to write their final WASSCE.

Figure 1 Conceptual Framework

International Journal of Engineering Technologies and Management Research 68

Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

3.1. DATASET AND STUDY AREA

The population (N) for the current study is two thousand four hundred (2,400)
students from Sunyani Senior High school, which constitute two thousand past
students and four hundred (400) continue students in SHS 3. The study adopted a
simple random sampling technique stated in algorithm 1 to select one thousand five
hundred (1,500) students (75%) out of 2,000 past students’ records and 5% (20)
out of the 400 continue students, making the sample size for the study to be 1520.
The 20 continue students were taken to make predictions against the pending 2020
WASSCE exams. The study records consist of student’s continuous assessment for
SHS 1 and SHS 2 (6 terms) and demographic data from students offering business,
general art, visual arts, science, and home economics courses.

Algorithm 1 simple random sampling

1. Make a list of classes (C) c = {c1 , c2 ,..., ct } where t is the total class, for this study t = 5
2. For each ci in c , where i = (1, 2,..., t )
3. Make a list of all the students (population) (N) in the class ( ci ) (the list contains their index
numbers)

4. Assign a sequential number to each student (1,2,3…n) in the class ( ci ) . This represents the
sampling frame (percentage of students for a class). The list from which the study draws its random
sample

5. Initialize the sample size (k) for the class ( ci ) , k =

( 0.75 × N ) or ( 0.05 × N ) for past
students and continue students’ class respectively

6. Select the sample (k) of ( ci ) using a random number generator (rand ()), based on the
sampling frame (population size of ( ci ) )

 / / Generates k random number within 1 – n. 

Initialise random number generator   rand (1 : N )
Initialise an array ( A ) of size ( k ) to hold generated numbers
For = int i 0 to ( k − 1)
    A [i ] =    rand .Next (1, N )
end

7. Associate every generated number in (A) with its (n) equivalent number and select the index
number of students in ( ci )
8. Repeat steps 3 to 7 for all ci in c

Training Techniques: The training technique adopted for the current study is a
supervised machine learning technique, where the intended input variables
(demographic and past terminal exams record) are entered into the network to
produce the required output variable (WASSCE score).

International Journal of Engineering Technologies and Management Research 69

Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

The RF is applied to the Train_D and the model learning the pattern hidden in
the dataset, we measure the accuracy and error metrics and determine if they are
within the accepted values. If they are, then the learned pattern is then applied to
the (Test_D) to make a prediction.
Random forest (RF): This is an ensemble-learning technique that integrates the
performance of several decision tree algorithms to predict or classify the value of a
variable Rodriguez-Galiano et al. (2015), Wakefield, 2013). RF can be used for both
classification and regression ML tasks. In the RF technique, many DTs are created,
with each observation fed into every single DT. The optimal result in each
observation is used as the final output. A different observation is served into all the
trees and a majority vote is computed for each classification model. An error
estimate is made for the cases, which were not used throughout the tree building.
This is known as OOB (Out-of-bag) error estimate, which is stated as a percentage.
When an RF receives an input of (x), where x is a vector consisting of variables of
different evidential features examined for a given training area, the RF builds
several regression trees (N) and averages the results. Therefore, that for N tress
{T(x)} ^N the RF regression predictor is given by equation Equation 1.

𝟏𝟏
𝒇𝒇𝑵𝑵 𝑵𝑵
𝒓𝒓𝒓𝒓 (𝒙𝒙) = ∑𝑵𝑵=𝟏𝟏 𝑻𝑻(𝒙𝒙) Equation 1
𝑵𝑵

Algorithm 2 Random Forest Algorithm

Input : DatasetTrain _ D, number of trees in the ensemble n

Output : a composite model M *
1 → for j = 1 to k do
2 → Construct boostrap sample Train Dj by sampling S with replacement
3 → Select 4 features randomly
4 → UseTrain _ D j and unsystematically selected 4 features to develop a regression tree M j
5 → end for
6 → return M*

TrainD = {( x1 , y1 ) , ( x2 , y2 ) ,..., ( xn , yn )} The Bootstrap Aggregating (Bagging)

technique was used in assembling the decision trees for this study. Given a Train_D
as given in equation Equation 2, the bagging technique generates a New_Di new
training dataset of size denoted by N which is sampled from the original training
dataset Train_D with replacement. New_Di is referred to as the bootstrap sample.
By bootstrapping, some observations may be recurrent in each New_Di. This
approach assists to reduce variance and circumvents overfitting. Users specify the
number of regression trees (T), in the current study, two hundred (200) trees were
specified.

TrainD = {(𝑥𝑥1 , 𝑦𝑦1 ), (𝑥𝑥2 , 𝑦𝑦2 ), , … . . , (𝑥𝑥𝑁𝑁 , 𝑦𝑦𝑁𝑁 ), } Equation 2

Choosing Variables to Split On: grow unpruned regression tree using the steps
for each of the bootstrap samples: At the individual node, indiscriminately sample
K variables and select the most exceptional split among those variables (K) rather
than picking the most excellent split amid all predictors. This practice is sometimes
International Journal of Engineering Technologies and Management Research 70
Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

called “feature bagging.” We select a random subset of the predictors or features

because the correlation of the trees in a standard bootstrap sample can be reduced.
In this study K=p⁄3.
The splitting principle, let assume that a partition is divided into T
constituencies R1, R2…, RT. We model the response as a constant ck in each
constituency as proposed by Wu et al. (2017) in equation Equation 3. The splitting
principle at each node is to minimize the sum of squares. Hence, the best c ̂t_tis the
average of yi in region Rt as given in equation Equation 4.

𝒇𝒇(𝒙𝒙) = ∑𝑻𝑻𝒕𝒕=𝟏𝟏 𝒄𝒄𝒕𝒕 𝑰𝑰(𝒙𝒙 ∈ 𝑹𝑹𝒕𝒕 ) Equation 3

(c_t) =
̂ ave (y_i│x_i∈R_t) Equation 4

Let a splitting feature j and split point s, and express the pair of half-planes

𝑅𝑅1 (𝑗𝑗, 𝑠𝑠) = �𝑋𝑋|𝑋𝑋𝑗𝑗 ≤ 𝑠𝑠� 𝑎𝑎𝑎𝑎𝑎𝑎 𝑅𝑅2 (𝑗𝑗, 𝑠𝑠) = �𝑋𝑋|𝑋𝑋𝑗𝑗 ≥ 𝑠𝑠� Equation 5

Where j and s satisfy equation Equation 6.

𝑚𝑚𝑚𝑚𝑚𝑚 𝑚𝑚𝑚𝑚𝑚𝑚
� ∑𝑥𝑥𝑖𝑖 ∈𝑅𝑅1(𝑗𝑗,𝑠𝑠)(𝑦𝑦𝑖𝑖 − 𝑐𝑐1)2 + ∑𝑥𝑥𝑖𝑖 ∈𝑅𝑅2(𝑗𝑗,𝑠𝑠)(𝑦𝑦𝑖𝑖 − 𝑐𝑐2)2 � Equation 6
𝑐𝑐 1 𝑐𝑐 2

When the best split is obtained, the dataset is partitioned into two resulting
segments and echo the splitting procedure on each of the two segments. This
splitting procedure is reiterated until a predefined ending criterion (threshold) is
satisfied, five was set as the threshold for this study.

3.2. EVALUATION METRICS

Several criteria are available for measuring the performance of a predictive
model; however, for this study, the following criteria were used: Root means
squared error (RMSE): This index evaluation the residual between the actual-value
and predicted-value. A model has better performance if it has a smaller RMSE. An
RMSE equal to zero represents a perfect fit.

1
𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 = � ∑𝑚𝑚
𝑣𝑣=1(𝑡𝑡𝑣𝑣 − 𝑦𝑦𝑣𝑣 )
2
Equation 7
𝑚𝑚

Mean absolute percentage error (MAPE): This index shows an average of the
absolute percentage errors, the lower the MAPE, the better.

International Journal of Engineering Technologies and Management Research 71

Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya
𝟏𝟏 𝒕𝒕 −𝒚𝒚𝒗𝒗
𝑴𝑴𝑴𝑴𝑴𝑴𝑴𝑴 = ∑𝒎𝒎
𝒗𝒗=𝟏𝟏 �
𝒗𝒗
� Equation 8
𝒎𝒎 𝒕𝒕𝒗𝒗

tv is the actual value, yv is the predicted value produced by the model, and m is
the total number of observations. The correlation coefficient (R): This criterion
reveals that the strength of associations between actual-values and predicted-
values. The correlation coefficient has a range from 0 to 1, and a model with a higher
R means it has better performance.

∑𝑚𝑚
𝑣𝑣=1(𝑡𝑡𝑣𝑣 −𝑡𝑡̅)(𝑦𝑦𝑣𝑣 −𝑦𝑦�)
𝑅𝑅 = Equation 9
�∑𝑣𝑣=1(𝑡𝑡𝑣𝑣 −𝑡𝑡̅)2 .∑𝑚𝑚
𝑚𝑚 (
𝑣𝑣=1 𝑣𝑣 −𝑦𝑦
𝑦𝑦 �)2

In addition to the mentioned criteria, output accuracy is also used to evaluate

the performance of the predictive model.

4. ANALYSIS OF RESULTS AND DISCUSSION

4.1. DEMOGRAPHIC CHARACTERISTICS OF THE STUDENTS
The background data of the students were sought. The data elicited relate to
gender, age distribution, and classes/forms. The background data of students
formed no part of the principal analysis. The reason for presenting the background
data of respondents was to have an idea about the over-all information of students
but not for the principal scrutiny. The results are discussed in the form of frequency
and percentages as shown in Table 2

Table 2 Demographic Features of the Students

Variable Sub-scale Frequency Percentage (%)

(n = 1520)
Gender Male 670 44.07
Female 850 55.9
Age Distribution 14 -15 Years 180 11.8
16 - 18Years 950 62.5
≥19 Years 390 25.7

Table 2 shows the students’ demographic information. It was observed that out
of one thousand five hundred and twenty (1520) students, a majority (n = 850;
55.9%) were females while 670 (44.07%) were males. This outcome shows that the
female students were 26% more than their male counterparts were. The majority
(n = 950; 55.9%) of the students were within the age group of 16 – 18 years, which
confirms the normal age for one to be in SHS, while 180 of them representing 11.8%
were found to be in the age group of 14-15 years and 390 representing (25.7%)
were found with 19 years or more.

4.2. DESCRIPTIVE STATISTICS

Table 3 shows the statistical analysis of candidate continuous assessment
scores. The mean value range (63.67 - 66.36) for all subjects except core maths
shows that the end of terms examination mark of the candidates was quite good.
International Journal of Engineering Technologies and Management Research 72
Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

However, the minimum score (0) and maximum score (80.02) for core maths
indicates that the performance of a candidate in core maths need more
improvement. This outcome affirms Ablakwa’s reports that a high percentage of
candidate fails in the core maths (Ablakwa, 2014).
Additional statistical analysis was performed on the students’ continuous
assessment scores. Table 4 shows a one-sample test of 95% confidence interval.
Again, the difference in mean values in students’ continuous assessment shows a
close margin in the general performance of students.

Table 3 Statistical Analysis of Student Continuous Assessment

Descriptive Statistics
N Minimum Maximum Mean Std. Deviation
CM 1520 0 80.02 66.2028 7.6342
EG 1520 44.33 81.2 66.3631 7.59182
IS 1520 46 84.25 63.672 6.46139
SS 1520 45.5 84.75 64.8148 7.96228
EL1 1520 39.17 87.6 63.8926 7.54045
EL2 1520 34.17 83 65.2059 6.88485
EL3 1520 46.5 84.8 66.0408 6.87008
EL4 1520 29.83 89.6 65.7409 7.50126
Valid N (listwise) 1520

Table 4 One-Sample Test

Test Value = 0
95% Confidence Interval of the
Difference
t df Sig. (2- Mean Lower Upper
tailed) diff.
CM 153.664 1520 0 66.20277 65.3551 67.0505
EG 174.377 1520 0 63.52882 62.812 64.2456
IS 144.143 1520 0 64.62707 63.7449 65.5092
SS 150.192 1520 0 63.70774 62.8731 64.5423
EL1 166.222 1520 0 65.05242 64.2824 65.8224
EL2 169.554 1520 0 65.97631 65.2107 66.7419
EL3 153.81 1520 0 65.6108 64.7715 66.4501
EL4 168.268 1520 0 65.36828 64.6039 66.1326

4.3. FEATURE IMPORTANCE RANKING

The correlation between each input feature and the expected output (WASSCE
score) was tested. Figure 2 shows the feature's importance ranking. The result
shows that the demographic data (age and gender) are of less significance. Thus, the
gender and age of a candidate do not determine how good or bad he/she will pass
the final WASSCE. Again, the program pursued by a candidate also is of less
importance to one passing the WASSCE. However, students’ performance in
Integrated Science (IS) and Social Studies (SS) were seen to be the most significant
determining factor of one passing his/her WASSCE. Furthermore, the results show
that the average continuous assessment score of Core Math (CM) by a candidate was

International Journal of Engineering Technologies and Management Research 73

Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

also highly related to the candidate success in the final WASSCE. This affirms the
reason behind core math being one of the compulsory subjects in all areas of studies.

Figure 2 Features Importance

4.4. MODEL TRAINING AND TESTING

Figure 3 shows the Out-of-bag error rate of the proposed random forest
regressor. The aim here was to obtain the right number of estimators that will offer
fewer errors. The minimum number of estimators was set to fifteen (15) and the
maximum to two hundred (200). The result shows that as the estimators increase
in number, the error margin reduces. However, the error increased when the
number of estimators increased above one hundred and seventy-five (175). Hence,
for this study, the number of estimators was set to one hundred and sixty (160).

Figure 3 Out-of-bag Training Error

International Journal of Engineering Technologies and Management Research 74

Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

4.5. ACCURACY AND ERROR METRICS

Table 5 shows the error metrics and computational time of the proposed
predictive model. With a training dataset of one thousand one hundred and twenty-
five (1,125) rows eleven (11) columns, it took the proposed model 0.392 seconds to
study all hidden patterns in the dataset. While 0.013 was taken to make predictions
on three hundred and seventy-five (375) test data. The RMSE of 0.001639 and the
MAE error of 0.001321 achieved by the proposed model show that the proposed
model fits very well with the dataset. Again, the obtained error values indicate a
close range in model estimated value compared with the actual (true) values and an
indication that students’ performance in the WASSCE can be effectively predicted
based on student performance in continuous assessment.

Table 5 Model error metrics and computational time

RMSE MAE R^2 Training time Testing Time

0.001639 0.001321 0.034323 0.392 secs 0.013 secs

Figure 4 shows the grade analysis based on model prediction against actuals on
the test dataset. It was observed that a good number of the students records at least
A1 in one or more of the subject grades for the final WASSCE. Also, the closeness of
the model predicted to the actual grade confirms the accuracy of the proposed
model (see Table 5 and 4.7). Based, on the outcome it can be inferred that a good
predictive model can help school authority to ascertain the grade analysis of their
students before the WAASCE exams.

Figure 4 Grade analysis of predicted values

To further examine the proposed predictive framework, the continuous

assessment records of 20 students that are in SHS 3 preparing for the final WASSCE
was used to predict the outcome of their exams. Table 6 shows the model predicted
values. Since the actual values were not available at the time of this study, there was
no means to compare the predicted values for these 20 students to their actual
International Journal of Engineering Technologies and Management Research 75
Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

results. However, based on the accuracy record by the model on the test data, it is
hoped that the predicted results will not be far from right when these students take
their final exams.

Table 6 Model predicted values for 20 students yet to write their WASSCE

Record WCM WEG WIS WSS WEL1 WEL2 WEL3 WEL4

Y' Y' Y' Y' Y' Y' Y' Y'
1. 6 5 5 4 4 4 4 4
2. 4 3 4 3 3 4 4 4
3. 4 4 4 3 4 4 4 5
4. 6 4 5 4 4 4 4 4
5. 5 4 5 4 5 5 4 4
6. 6 5 5 4 4 4 4 4
7. 6 5 6 4 4 4 5 5
8. 3 4 4 3 4 4 4 5
9. 4 4 4 3 4 4 4 4
10. 3 4 4 4 4 5 4 4
11. 4 4 4 3 4 4 4 4
12. 3 4 4 4 3 3 4 3
13. 5 5 5 4 4 4 4 4
14. 5 4 5 4 4 4 4 4
15. 5 4 5 3 4 4 4 4
16. 4 4 5 3 3 4 5 6
17. 4 4 4 3 4 4 4 5
18. 4 4 4 3 4 4 4 4
19. 4 4 4 3 4 4 4 4
20. 5 4 5 4 4 4 4 4

4.6. COMPARISON ANALYSIS OF RESULTS WITH RELATED

WORKS
Table 7 shows a comparative analysis of the current study with related works.
Pandey and Taruna (2016) obtained an accuracy of (86.76 – 98.86) % with 1000
respondents, Guo et al. (2015) an accuracy (63.2 – 88.4) % with 120000
participants. The results as shown in Table 7, indicates that the prediction accuracy
(89.4%) with 1000 participant outperformed other state-of-the-art students on
student’s academic performance. This comparative analysis shows that there has
been an enhancement in the current study to existing literature and a confirmation
that the academic performance of students can effectively be predicted with a high
accuracy rate.

Table 7 A comparative analysis of study results with related works

References Predicted variable Sample ML algorithm Evaluation metrics.

size
Accuracy RMSE
Purwaningsih English score in 72 Naïve Bayes 35-86% Not
et al. (2018) WASSCE Available
Bosson- Core Mathematics 42 Linear regression 90 – 92 % Not
Amedenu score in WASSCE Available
(2017)

International Journal of Engineering Technologies and Management Research 76

Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

Ahmed and Not Stated 1547 Decision tree Not Not

Ibrahim (2014) (ID3) Available Available
Pandey and Not Stated 1000 Decision Tree, K- 86.76 – Not
Taruna (2016) Nearest 98.86 % Available
Neighbour, and
Aggregating One-
Dependence
Estimators
(AODE)
Livieris et al. Mathematics 279 Decision trees, 51.6 – Not
(2016) Support vector 70.3 % Available
machine and
Sequential
Minimal
Optimization
(SMO) algorithm
Li et al. (2013) Not stated 90 Principal Not Not
Component Available Available
Analysis
Musso et al. Psychology, 786 Artificial Neural 87 – Not
(2013) engineering, Networks 100% Available
medicine, law, social
communication,
business, and
marketing
Guo et al. Not stated 120,000 Deep neural 63.2 – Not
(2015) networks 88.4% Available
Osmanbegovic 257 Naive Bayes 71.2 – 0.42 – 0.5
and Suljic algorithm, 76.65%
(2012) Multilayer
Perceptron and
Decision tree
Osmanbegović Not stated 1410 Random Forest 71% Not
et al. (2014) and Decision Available
Trees (J48)
Berhanu Not stated 199 Decision trees 84.95% Not
(2015) Available
Tran et al. Not stated 1268 ANN, DT and Not 1.668
(2017) Support Vector Available
Machines (SVMs)
Current study General Science 1,520 Random Forest 89.40% 0.001639
(Adjei 2020) Business
Home Economics
Visual Arts
General Arts

5. CONCLUSIONS AND RECOMMENDATIONS

The study aimed at investigating the factors that affect student academic
performance in the WASSCE. It also attempted to predict the final students' score in
the WASSCE based on student’s continuous assessment using the random forest
algorithm. The considered factors were demographic factors such as age and
gender, and continuous assessment score. Descriptive statistical (frequency and
percentages, means, and standard deviation) and T-Test methods were used to
analyse the details of one thousand (1000) randomly sampled students’ records
from the Education Database (ED). The study found that the demographic data
International Journal of Engineering Technologies and Management Research 77
Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

(gender and age) has no effects on the performance of students in the WASSCE. A
high correlation was found between Integrated Science (SI) and Social Studies (SS),
and the performance of students in the WASSCE. The study found that students that
score within (40-55) in core maths were likely not to perform well in the final
WASSCE. The performance of students in the WASSCE is highly predictable at an
accuracy of 89.4%.

ACKNOWLEDGEMENTS
People who contributed to the work but do not fit criteria for authorship should
be listed in the Acknowledgments, along with their contributions. It is advised that
authors ensure that anyone named in the acknowledgments agrees to being so
named. Funding sources that have supported the work should also be cited.

REFERENCES
Adejo, O. W., & Connolly, T. (2018). Predicting Student Academic Performance Using
Multi-Model Heterogeneous Ensemble Approach. Journal Of Applied
Research In Higher Education, 10(1), 61-75. Retrieved from
[Link]
Agrawal, H., & Mavani, H. (2015). Student Performance Prediction Using Machine
Learning. 4(03), 111-113. Retrieved from
[Link]
Ahmed, A. B. E. D., & Ibrahim, S. E. (2014). Data Mining : A Prediction For Student's
Performance Using Classification Method. World Journal Of Computer
Application And Technology, 2(2), 43-47. Retrieved from
[Link]
[Link]. (2010). Random Forest Algorithm. Retrieved from
Https://[Link]/Imgres?Imgurl=Https%3A%2F%[Link]
[Link]%2Fwp-Content%2Fuploads%2F2015%2F06%2Frandom-
[Link]&Imgrefurl=Https%3A%2F%[Link]%2
Fblog%2F2015%2F06%2Ftuning-Random-Forest-
Model%2F&Tbnid=Bldygobmf_Oqom&Vet=12ahukewibtpbop-
_Qahvtlkqkhbfndnuqmygieguiarc3aq.I&Docid=Gp-
Attuquayefio, Niiboi, S., & Addo, H. (2014). Using The UTAUT Model To Analyze
Students' ICT Adoption. International Journal Of Education & Development
Using Information & Communication Technology, 10(3), 75-86. Retrieved
from
Http://[Link]/Login?Url=Http://[Link]/Logi
[Link]?Direct=True&Db=Ehh&AN=97923459&Site=Ehost-Live
Berhanu, F. (2015). Students ' Performance Prediction Based On Their Academic
Record. International Journal Of Computer Applications, 131(5), 27-35.
Retrieved from [Link]
Bhardwaj, B. K., & Pal, S. (2011). Data Mining : A Prediction For Performance
Improvement Using Classification. International Journal Of Computer
Science And Information Security, 9(4), 136-140.
Bosson-Amedenu, S. (2017). Predictive Validity Of Mathematics Mock Examination
Results Of Senior And Junior High School Students' Performance In WASSCE
International Journal of Engineering Technologies and Management Research 78
Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

And BECE In Ghana. Asian Research Journal Of Mathematics, 3(4), 1-8.

Retrieved from [Link]
Chen, J.-F., Hsieh, H.-N., & Quang, H. Do. (2014). Predicting Student Academic
Performance : A Comparison Of Two Meta-Heuristic Algorithms Inspired By
Cuckoo Birds For Training Neural Networks. Algorithms, 7(4), 538-553.
Retrieved from [Link]
Cortez, P., & Silva, A. (2008). Using Data Mining To Predict Secondary School Student
Performance. 5th Annual Future Business Technology Conference,
2003(2000), 5-12. Retrieved from
Https://[Link]/10.13140/RG.2.1.1465.8328
Devasia, T., Vinushree, T. P., & Hegde, V. (2016). Prediction Of Students Performance
Using Educational Data Mining. En 16. Retrieved from
[Link]
Egbenya, G. R. K., & Halm, E. (2016). A Comparative Study Of Students ' Performance
For The Three Year And The Four Year Programmes In Mfantsipim And
University Of Cape Coast, Practice Senior High Schools In Cape Coast , Ghana.
Internal Journal Of Innovative Reserach & Development, 5(3), 114-129.
Fleischer, J. E. (2015). Information Communication Technology Usage Patterns In
Second Cycle Schools : A Study Of Two Selected Senior High Schools In Ghana
(Issue 10232161).
Goga, M., Kuyoro, S., & Goga, N. (2015). A Recommender For Improving The Student
Academic Performance. Procedia - Social And Behavioral Sciences,
180(November 2014), 1481-1488. Retrieved from
[Link]
Guo, B., Zhang, R., Xu, G., Shi, C., & Yang, L. (2015). Predicting Students Performance
In Educational Data Mining. 2015 International Symposium On Educational
Technology (ISET), 125-128. Retrieved from
[Link]
Hedén, W. (2016). Predicting Hourly Residential Energy Consumption Using
Random Forest And Support Vector Regression An Analysis Of The Impact
Of Household Clustering On The Performance Accuracy. KTH Royal Institute
Of Technology.
Khasanah, A. U., & Harwati. (2017). A Comparative Study To Predict Student's
Performance Using Educational Data Mining Techniques. IOP Conference
Series: Materials Science And Engineering, 215(1). Retrieved from
[Link]
Kieti, J. M. (2017). An Investigation Into Factors Influencing Students' Academic
Performance In Public Secondary Schools In Matungulu Sub-County,
Machakos County. In South Eastern Kenya University. Retrieved from
[Link]
Kumar, M., & Thenmozhi, M. (2006). Forecasting Stock Index Movement: A
Comparison Of Support Vector Machines And Random Forest. Indian
Institute Of Capital Markets 9th Capital Markets Conference Paper, 1-16.
Retrieved from [Link]

International Journal of Engineering Technologies and Management Research 79

Adjei-Pokuaa Henrietta, and Adebayo F. Adekoya

Leo, B., & Adele, C. (2002). Random Forests. Retrieved from

Https://[Link]/~Breiman/Randomforests/Cc_Home.Ht
m
Li, K. F., Rusk, D., & Song, F. (2013). Predicting Student Academic Performance. 2013
Seventh International Conference On Complex, Intelligent, And Software
Intensive Systems, 27-33. Retrieved from
[Link]
Livieris, I. E., Tassos, A. M., & Panagiotis, P. (2016). A Decision Support System For
Predicting Student Performance. International Journal Of Innovative
Research In Computer And Communication Engineering, 02(12), 7232-
7237. Retrieved from [Link]
Mahbina, A. M., & Zamil, K. M. S. (2018). GIS-Based Analysis Of Changing Surface
Water In Rajshahi City Corporation Area Using Support Vector Machine
(SVM), Decision Tree & Random Forest Technique. Machine Learning
Research, 3(2), 11-17. Retrieved from
[Link]
Mohamed, A., Rizaner, A., & Hakan, A. (2016). Using Data Mining To Predict
Instructor Performance. Procedia - Procedia Computer Science,
102(August), 137-142. Retrieved from
[Link]
Musso, M. F., Kyndt, E., Cascallar, E. C., & Dochy, F. (2013). Predicting General
Academic Performance And Identifying The Differential Contribution Of
Participating Variables Using Artificial Neural Networks. Frontline Learning
Research, 1(1), 42-71. Retrieved from [Link]
Nti, I. K., Adekoya, A. F., & Weyori, B. A. (2020). A Comprehensive Evaluation Of
Ensemble Learning For Stock-Market Prediction. Journal Of Big Data, 7(1).
Retrieved from [Link]
Oladokun, V. O., Adebanjo, A. T., & Charles-Owaba, O. E. (2008). Predicting Students'
Academic Performance Using Artificial Neural Network : A Case Study Of An
Engineering Course. The Pacific Journal Of Science And Technology, 9(1),
72-79.
Osmanbegovic, E., & Suljic, M. (2012). Data Mining Approach For Predicting Student
Performance. Journal Of Economics And Business, X (1), 3-12.
Osmanbegović, E., Suljić, M., & Agić, H. (2014). Determining Dominant Factor For
Students Performance Prediction By Using Data Mining. Vitez-Tuzla-Zagreb-
Beograd-Bucharest, XVII (34), 147-158.
Pandey, M., & Taruna, S. (2016). Towards The Integration Of Multiple Classifier
Pertaining To The Student ' S Performance Prediction. Perspectives In
Science, 8, 364-366. Retrieved from
[Link]
Purwaningsih, N., Arief, D. R., Purwaningsih, N., & Arief, D. R. (2018). Predicting
Students ' Performance In English Class. AIP, 020020. Retrieved from
[Link]
Rajesh, S. B. (2018). Introduction To Decision Trees.
Https://[Link]/Greyatom/Decision-Trees-A-Simple-Way-To-
Visualize-A-Decision-Dc506a403aeb#:~:Text=A Decision Tree Is A
Flowchart-Like Structure In Which,Taken After Computing All Attributes).
International Journal of Engineering Technologies and Management Research 80
Predictive Analytics of Academic Performance of Senior High School (Shs) Students: A Case Study of Sunyani Shs

Rodriguez-Galiano, V., Sanchez-Castillo, M., Chica-Olmo, M., & Chica-Rivas, M.

(2015). Machine Learning Predictive Models For Mineral Prospectivity: An
Evaluation Of Neural Networks, Random Forest, Regression Trees And
Support Vector Machines. Ore Geology Reviews, 71, 804-818. Retrieved
from [Link]
Tan, Z., Yan, Z., & Zhu, G. (2019). Stock Selection With Random Forest : An
Exploitation Of Excess Return In The Chinese Stock Market. Heliyon, 5(8),
E02310. Retrieved from [Link]
Tran, T., Dang, H., Dinh, V., Truong, T.-M.-N., Vuong, T., & Phan, X. (2017).
Performance Prediction For Students : A Multi-Strategy Approach.
CYBERNETICS AND INFORMATION TECHNOLOGIES, 17(2), 164-182.
Retrieved from [Link]
Tuen, W., Leung, V., Yee, T., Pan, W., Wu, C., Lung, S. C., & Spengler, J. D. (2019).
Landscape And Urban Planning How Is Environmental Greenness Related
To Students ' Academic Performance In English And Mathematics ?
Landscape And Urban Planning, 181(1), 118-124. Retrieved from
[Link]
Wu, D., Jennings, C., Terpenny, J., Gao, R. X., & Kumara, S. (2017). A Comparative
Study On Machine Learning Algorithms For Smart Manufacturing: Tool
Wear Prediction Using Random Forests. Journal Of Manufacturing Science
And Engineering, 139(7), 071018. Retrieved from
[Link]
Yadav, S. K., & Pal, S. (2012). Data Mining : A Prediction For Performance
Improvement Of Engineering Students Using Classification. World Of
Computer Science And Information Technology Journal WCSIT, 2(2), 51-56.
Retrieved from [Link]
Yeboah, Y. K. (2014). Investigating The Low Performance Of Students' English In The
Basic Education Certificate Examination In The Sunyani Municipality (Issue
10357198) [UNIVERSITY OF GHANA, LEGON]. Retrieved from
[Link]
Yusif, H. M., Yussof, I., & Noor, A. H. S. M. (2011). Determinants Of Students Academic
Perform- Ance In Senior High Schools : A Binary Logit Approach. 31(3), 107-
117. Retrieved from [Link]

International Journal of Engineering Technologies and Management Research 81