COMPARATIVE ANALYSIS FOR DEEP LEARNING ALGORITHMS FOR

STUDENT PERFORMANCE PREDICTION.

CHAPTER ONE: INTRODUCTION

1.1 Background of the Study

Predicting student performance is an important task for educators and institutions to

understand how different factors influence students' academic success. Machine learning

algorithms, including Decision Tree (DT), k-Nearest Neighbors (KNN), and Support Vector

Machines (SVM), have been widely used in this domain. In this analysis, we compare these

algorithms based on their strengths and weaknesses for student performance prediction.

The rapid development of information technology (IT) has greatly increased the amount of

data in different institutions. Huge warehouses contain a wealth of data and constitute a

valuable information goldmine (Hashim et al., 2020). This dramatic inflation in the amount of

data in institutions has not kept pace with the efficient ways of investing these data. Thus, a

new challenge has recently emerged, that is, transitioning from traditional databases that store

and search for information only through questions asked by a researcher to techniques used in

extracting knowledge by exploring prevailing patterns of data for decision making, planning

and future vision (Hasan et al., 2019).

Every year, educational centers, schools, universities, and institutes admit many students in

different fields with various grades and performance capability. Recently in educational areas

for studying and teaching performance enhancement, data mining and other techniques

become well known. With huge amount of data in diverse technological areas, and generating

such kinds of data rapidly, there is a needs for proper usage (Rastrollo-Guerrero et al., 2020).

During the last few years, the application of artificial intelligence in education has grown

exponentially, spurred by the fact that it allows the discovery of new, interesting and useful
knowledge about students (Chitti et al., 2020). Advising students on their class performance

and motivating them in other to improve on their performance is an integral part of every

institution. Improving student’s academic performance is not an easy task for the academic

community of higher learning. As students’ behaviors are important factors that can reflect

their learning styles and living habits in institution, extracting useful features of them plays a

helpful role in understanding the students’ learning process, which is an important step

towards personalized education (Albreiki et al., 2021).

Forecasting student performance is essential for educators to obtain early feedback and take

immediate action or early precautions if necessary to improve the student’s performance

(Mengash, 2020). This prediction can be managed by locating the source of the problem.

Should it be from extra activities that the student is participating in, family problems, or

health problems? All these factors can have a major effect on student performance (Sathe &

Adamuthe, 2021). There is often a great need to be able to predict future students’ behavior in

order to improve curriculum design and plan interventions for academic support and guidance

on the curriculum offered to the students.

Student academic performance serves as a metric through which individuals can track their

academic progress, leading to certification. Various types of evaluations, including oral

presentations and practical training, contribute to assessing a student’s level of academic

competence. In a broader sense, the measurement of student performance relies on chosen

objectives and criteria, utilizing diverse standards such as GPA (Grade Point Average). The

GPA is the basis for studies on academic performance, which used as a tool to simplify a set

of chores, solve a collection of problems or as a tool to help with decision-making(Laakel

Hemdanou et al., 2024).

Elementary school academic performance evaluations determine a student’s educational path

and future possibilities. Developing models that predict elementary children’s academic

performance be comes increasingly important as policymakers and educators work to

improve the learning process and offer corrective measures. More sophisticated techniques

are required since conventional methods of assessment and prediction fail to capture the

complicated dynamics of a student’s academic career. An aspect of artificial intelligence

called deep learning has demonstrated incredible ability in several fields of study. Its use to

predict academic achievement in elementary school kids provides the possibility of more

precise and insights based on data(Deng et al., 2024). Hence, the multi-class models need

more studies to improve prediction performance. As techniques used in the prediction of

student performance in online higher education, traditional artificial intelligence techniques

are commonly applied while more advanced techniques like deep learning are rarely applied.

Since the majority of studies have taken the approach of developing predictive models that

target specific courses, but overfitting can take place if new courses are devised(Al-azazi &

Ghurab, 2023).

Thus, analyzing and processing these data carefully can give a useful information about the

students’ knowledge and the relationship between them and the academic tasks. This

information is the source that feeds promising algorithms and methods, to be able to predict

students’ performance. Hence, this research work proposed a comparison evaluation of the

following supervised learning algorithms; Decision Tree (DT), k-Nearest Neighbors (KNN),

and Support Vector Machines (SVM) for student performance predictions. In other, to know

the best fit model among the three algorithms.

1.2 Problem Statement

Predicting student performance is a critical task for educators and institutions as it provides

valuable insights into the factors influencing academic success, allowing for timely

interventions to support students. With the rapid growth of information technology (IT),

educational institutions are generating vast amounts of data, which, if effectively analyzed,

can be harnessed to improve decision-making, curriculum design, and academic support.

However, the sheer volume of data has outpaced the development of efficient tools for its

analysis, creating a need to transition from traditional data storage and retrieval methods to

more advanced techniques that can extract meaningful knowledge from prevailing patterns in

the data (Hasan et al., 2019).

Deep learning algorithms, such as Decision Tree (DT), k-Nearest Neighbors (KNN), and

Support Vector Machines (SVM), have gained significant attention for their potential in

predicting student performance by uncovering complex relationships in student data.

However, comparative studies that systematically evaluate these algorithms in the context of

educational data are limited. Existing predictive models often suffer from challenges such as

overfitting, lack of generalizability, and difficulties in balancing accuracy with

interpretability (Sathe & Adamuthe, 2021). These issues are exacerbated in diverse

educational environments, where student demographics, curricula, and external factors vary

significantly.

Traditional assessment methods, such as Grade Point Average (GPA), provide only a narrow

view of student performance and fail to account for the multifaceted factors influencing

academic success, such as behavioral patterns, health issues, or family challenges (Laakel

Hemdanou et al., 2024). Furthermore, predicting student performance is essential for

identifying at-risk students early and providing tailored interventions to improve learning

outcomes. However, current models do not sufficiently address the complexities of academic
performance across different educational contexts, particularly in online learning and higher

education environments (Al-azazi & Ghurab, 2023).

This study aims to fill the gap in the existing literature by comparing the effectiveness of

three widely used deep learning algorithms—Decision Tree (DT), k-Nearest Neighbors

(KNN), and Support Vector Machines (SVM)—in predicting student performance. By

analyzing the strengths and weaknesses of these models, the research seeks to identify the

most suitable algorithm for predicting student success across various educational settings.

The findings will contribute to the development of more robust, interpretable, and

generalizable predictive models, which can enhance personalized learning interventions and

ultimately improve student outcomes (Mengash, 2020; Albreiki et al., 2021).

1.3 Aim and objectives of the study

The aim of this research work is to develop and compare the following deep learning

algorithms; Decision Tree (DT), k-Nearest Neighbors (KNN), and Support Vector Machines

(SVM) for student performance prediction and determine the best fit model among them.

This can be achieved through the following objectives.

i. To developed and train models using the following deep learning algorithms;

Decision Tree (DT), k-Nearest Neighbors (KNN), and Support Vector Machines

(SVM) for student performance prediction.

ii. To compared the developed models and determine the best fit model among them.

1.4 Scope of the Study

This study focuses on the development and comparison of three deep learning algorithms—

Decision Tree (DT), k-Nearest Neighbors (KNN), and Support Vector Machines (SVM)—for

predicting student performance in educational settings. The scope of this research is confined
to the application of these algorithms to datasets that contain information on various factors

influencing student performance, such as academic behaviors, socio-demographic factors,

extracurricular involvement, and personal circumstances.

The study will address the following key aspects:

i. The research will involve the development and training of predictive models using

Decision Tree (DT), k-Nearest Neighbors (KNN), and Support Vector Machines

(SVM). These models will be built using student data available from educational

institutions, which may include data such as grades, attendance, study habits, and

other relevant variables.

ii. A major focus of this study is the comparative evaluation of the three algorithms in

terms of their ability to predict student performance accurately and reliably. The study

will compare the strengths and weaknesses of each algorithm, particularly in terms of

accuracy, interpretability, generalizability, and susceptibility to overfitting.

iii. The study will primarily focus on traditional classroom settings and Virtual Learning

Environments (VLEs), comparing student performance across different educational

contexts. This will help in understanding how well the algorithms generalize to

different student populations, curricula, and learning environments, particularly with

respect to factors such as student demographics and external challenges.

iv. The evaluation of the models will use key performance metrics such as accuracy,

precision, recall, F1 score, and confusion matrix analysis. These metrics will provide

insights into the predictive power of each model, particularly in identifying students at

risk of underperforming.

v. The findings of this research will contribute to the development of predictive models

that can be applied in various educational settings to improve decision-making,

 curriculum design, and personalized learning interventions. The study aims to provide

educators with insights that can help them identify at-risk students early, design

effective interventions, and improve overall academic outcomes.

1.5 Limitation of the Study

The accuracy and effectiveness of the deep learning models developed in this study are

heavily dependent on the quality and comprehensiveness of the student data available. The

study will rely on existing datasets that may be limited in scope or completeness, potentially

impacting the robustness of the results. Missing data, inconsistent records, or inaccuracies in

the data could introduce biases or reduce the generalizability of the models. The study is

limited to the analysis of three specific deep learning algorithms—Decision Tree (DT), k-

Nearest Neighbors (KNN), and Support Vector Machines (SVM)—and may not account for

other deep learning techniques, such as deep learning models or ensemble methods, which

could offer improved prediction performance. As such, the findings of the study may not be

applicable to other deep learning algorithms or advanced techniques not explored in this

research.

The performance of the deep learning algorithms depends heavily on the features selected

from the available data. This study will be limited to the features included in the dataset and

may not explore the potential impact of additional features or more sophisticated data

preprocessing techniques. Moreover, the choice of feature engineering methods and data

transformations could influence model performance, limiting the potential for exploring more

advanced or customized approaches. Each of the deep learning algorithms used in this study

has inherent strengths and weaknesses. For instance, Decision Trees may suffer from

overfitting when not properly tuned, k-Nearest Neighbors may struggle with high-

dimensional data, and Support Vector Machines require careful tuning of hyperparameters for
optimal performance. These limitations may impact the results, and the study may not

explore all possible ways to mitigate these challenges (e.g., hyperparameter tuning, feature

scaling, or model ensembling). The dataset used in this study may not capture the full

diversity of student populations, which can vary significantly across different regions,

cultural backgrounds, and educational systems. As a result, the predictive models may not

generalize well to different demographic groups or educational contexts, limiting the

applicability of the findings to specific populations.

Student performance is influenced by a wide range of factors beyond academic and

behavioral data, such as health conditions, family issues, and socio-economic background.

This study will focus on the data available within the datasets, which may not fully account

for these external factors. Consequently, the predictive models may not capture the full

complexity of student performance and may overlook important contextual variables.

Although efforts will be made to avoid overfitting, the nature of deep learning models,

especially when trained on small or imbalanced datasets, may still lead to overfitting issues.

Additionally, while Decision Trees are relatively interpretable, k-Nearest Neighbors and

Support Vector Machines can be more challenging to interpret, which may hinder their

practical application in real-world educational settings where explainability is important for

educators and administrators. The study will evaluate the models using standard performance

metrics such as accuracy, precision, recall, and F1-score. However, there are other

performance metrics that may provide deeper insights into the effectiveness of the models,

such as ROC curves, AUC scores, or cost-sensitive evaluation. The use of a limited set of

metrics may restrict the comprehensive evaluation of the models’ performance across

different contexts.

1.6 Justification of the Study

Predicting student performance has become an essential tool for educators, institutions, and

policymakers aiming to enhance academic outcomes and provide personalized support to

students. This study is justified by several key factors that highlight the importance of

exploring deep learning algorithms in the context of student performance prediction. With the

rapid advancement of information technology (IT), educational institutions are generating

vast amounts of data. This data, if effectively analyzed, has the potential to unlock valuable

insights about student behaviors, learning patterns, and academic performance. However, the

traditional methods of analyzing this data, such as GPA and basic assessments, are

increasingly inadequate in capturing the complexity of student success. Deep learning

techniques provide a more efficient and effective approach to harnessing this data and

predicting student performance, making it a critical area of research. Traditional academic

performance metrics, such as Grade Point Average (GPA), fail to account for the myriad of

factors influencing student success, such as behavioral patterns, extracurricular activities,

family situations, and health issues. These methods provide a narrow view of performance

and cannot predict future outcomes with high accuracy. By applying machine learning

algorithms to student data, this study aims to develop more nuanced and accurate models that

can predict student performance and identify at-risk students early, facilitating timely

interventions and support. While deep learning algorithms such as Decision Trees (DT), k-

Nearest Neighbors (KNN), and Support Vector Machines (SVM) have been applied in

educational contexts, there is a lack of comprehensive studies that systematically compare

these algorithms based on their strengths and weaknesses for student performance prediction.

This study aims to fill this gap by evaluating and comparing these three popular algorithms,

offering valuable insights into which one is best suited for different educational

environments. By systematically comparing these models, the study will provide a clearer

understanding of their applicability in real-world settings.

The ability to accurately predict student performance allows educational institutions to make

more informed decisions regarding course offerings, teaching strategies, and academic

support services. By identifying patterns and correlations in student data, the study will

provide insights that can enhance decision-making processes at the institutional level,

ensuring that academic programs are aligned with students' needs and enhancing overall

institutional effectiveness. The comparative analysis conducted in this study will not only

identify the most effective algorithm for student performance prediction but will also offer

insights into how these algorithms can be adapted and scaled to different educational settings.

Whether in online learning environments, traditional classrooms, or higher education

institutions, the findings from this study can be applied to a wide range of contexts, providing

educators with a flexible tool that can be customized to suit diverse student populations and

institutional needs.

Educational data mining (EDM) is a rapidly evolving field that seeks to apply data-driven

techniques to improve education systems. This study will contribute to the field by evaluating

the effectiveness of specific deep learning algorithms in predicting student performance,

providing a foundation for further research in this area. By advancing the understanding of

how machine learning models can be utilized in educational contexts, the study will

contribute to the broader goal of using data to improve educational outcomes on a global

scale. Despite the growing interest in applying deep learning to education, there remains a

lack of research focused on comparative evaluations of commonly used algorithms in

predicting student performance. This study will fill this research gap, offering a

comprehensive comparison of three widely used algorithms (DT, KNN, and SVM) and

helping to inform future research and the development of more effective predictive models in

education.
1.7 Definition of Terms

i. Academic Performance:

Refers to the measurable outcomes of a student's academic activities, often evaluated

through grades, test scores, and other indicators of learning success (Chaurasia & Pal,

2020).

ii. Machine Learning (ML):

A branch of artificial intelligence that uses algorithms to analyze data, identify

patterns, and make predictions or decisions without being explicitly programmed

(Shahiri et al., 2015).

iii. Decision Tree (DT):

A supervised deep learning algorithm that models decisions and their possible

outcomes in a tree-like structure. It is widely used for its simplicity and

interpretability in classification and regression tasks (Romero & Ventura, 2020).

iv. k-Nearest Neighbors (KNN):

A non-parametric machine learning algorithm used for classification and regression. It

predicts the output of a new data point based on the 'k' closest data points in the

feature space (Kotsiantis et al., 2007).

v. Support Vector Machines (SVM):

A supervised learning algorithm that finds the optimal hyperplane to separate data

points of different classes, maximizing the margin between them for improved

classification and regression tasks (Zhang et al., 2016).

vi. Predictive Model:

A statistical or machine learning model designed to forecast future outcomes by

analyzing current and historical data trends (Shahiri et al., 2015).

vii. Overfitting:

A scenario in machine learning where a model learns the training data too well,

including noise and irrelevant details, leading to poor performance on new, unseen

data (Kotsiantis et al., 2007).

viii. Virtual Learning Environment (VLE):

An online platform that supports teaching and learning by offering tools for content

delivery, interaction, and course management (Romero & Ventura, 2020).

ix. Educational Contexts:

Refers to the various settings in which learning takes place, including traditional

classrooms, online environments, and hybrid models (Romero & Ventura, 2020).

x. Demographic Data:

Information about a population’s characteristics, such as age, gender, socioeconomic

status, and other relevant attributes, used to analyze their impact on academic

performance (Shahiri et al., 2015).

xi. Generalizability:

The extent to which a machine learning model or study findings can be applied to new

datasets, educational settings, or populations beyond the original scope (Zhang et al.,

2016).

xii. Computational Efficiency:

A measure of how effectively a machine learning algorithm processes data and

delivers results, considering the resources (time and memory) required (Kotsiantis et

al., 2007).

xiii. Interpretability:

The ability of a machine learning model to provide human-understandable

explanations for its predictions or decisions (Shahiri et al., 2015).

xiv. Educational Interventions:

Targeted strategies or actions designed to improve student learning outcomes and

address specific academic challenges (Zhang et al., 2016).

xv. Student Success:

A broad measure of a student's achievements in academic, social, and personal

development, often tied to the attainment of educational goals (Chaurasia & Pal,

2020).

1.8 ORGANIZATION OF THE PROJECT

This project is systematically organized into five chapters to ensure clarity and coherence in

presenting the study’s objectives, methods, findings, and conclusions. Below is an overview

of the structure:

Chapter One: Introduction

This chapter provides an overview of the study, including the background, problem

statement, aim, and objectives. It also outlines the scope, limitations, justification, and

significance of the study. Key definitions of terms are also included to ensure conceptual

clarity.

Chapter Two: Literature Review

This chapter reviews relevant literature on student performance prediction and machine

learning techniques. It covers the theoretical foundations of machine learning, an overview of

algorithms such as Decision Tree (DT), k-Nearest Neighbors (KNN), and Support Vector

Machines (SVM), and their applications in educational research. Additionally, this chapter

identifies research gaps and highlights the need for comparative analysis of these algorithms

in predicting academic performance.

Chapter Three: Methodology

The methodology chapter details the research design and methods used to achieve the study’s

objectives. It describes the datasets used, including the academic and demographic variables,

and the preprocessing techniques applied. The chapter also explains the implementation of

DT, KNN, and SVM algorithms, their evaluation metrics (e.g., accuracy, interpretability,

computational efficiency), and the comparative analysis framework. Ethical considerations

and limitations of the methodology are also discussed.

Chapter Four: Results and Discussion

This chapter presents the findings of the study, including the predictive performance of each

algorithm on the datasets analyzed. The strengths and weaknesses of DT, KNN, and SVM are

compared in terms of accuracy, interpretability, computational efficiency, and

generalizability. The results are discussed in relation to existing literature, and the

implications for educational practices and predictive modeling are explored.

Chapter Five: Summary, Conclusion and Recommendations

The final chapter summarizes the key findings of the study and draws conclusions based on

the evaluation of the machine learning algorithms. Recommendations are provided for the

most suitable algorithm(s) for predicting student performance in various educational settings.

This chapter also outlines suggestions for future research, including the exploration of

advanced techniques such as ensemble models, deep learning, and hybrid approaches.
 CHAPTER TWO: LITERATURE REVIEW

2.1 Introduction

This chapter aims to synthesize key findings from previous research, focusing on the

application of Decision Trees (DT), k-Nearest Neighbors (KNN), and Support Vector

Machines (SVM) for predicting student performance, as well as exploring the challenges and

opportunities these algorithms present. The review begins by exploring the role of data in

education, discussing the rapid increase in the volume and complexity of student-related data

due to advances in information technology. It then examines how deep learning models are

being utilized to analyze this data and provide predictive insights into student performance.

Specifically, it highlights the strengths and limitations of the three deep learning algorithms

under investigation in this study—DT, KNN, and SVM—drawing on a wide range of studies

that have applied these techniques to educational data.

2.2 Empirical Review

According to (Deng et al., 2024) It addresses that the essential need for predictive algorithms

to analyze and forecast academic results. Self-esteem and identity, as diverse psychological

variables, have been highlighted as crucial determinants in creating academic success. The

fundamental goal is to develop an effective prediction model for elementary school children’s

academic performance by including self-esteem and individuality in the analytical

framework. The architecture and parameters of the suggested approaches are optimized using

the OSO algorithm to provide a prediction model that remains accurate and efficient. The

OSO-DDNN is an essential instrument for educators and policymakers in identifying at-risk

pupils and implementing targeted interventions, supporting a comprehensive approach to

education that combines psychological well-being with academic performance.

Meanwhile, Learning Analytics aims to discover the class of students’ performance over

time. This helps instructors make in-time interventions but, discovering the students’

performance class in virtual learning environments consider a challenge due to distance

constraints. Many studies, which applied to Massive Open Online Courses (MOOC) datasets,

built predictive models but, these models were applied to specific courses and students and

classify students into binary classes. Additionally, the results of ANN-LSTM were compared

with the state-of-the-art models in terms of accuracy. The results show that the ANN-LSTM

model obtained the best results among baseline models (Al-azazi & Ghurab, 2023).

However, (C. Wang et al., 2024) said It is challenging for teachers to monitor each student’s

emotional state in real-time, making personalized learning difficult to achieve. Previous

emotion recognition methods, such as support vector machines, are limited by technology and

fail to meet practical application requirements. However, the development of deep learning

technology offers new solutions for facial expression recognition, which makes emotional

interaction and personalized support in education possible. Until now, there has been a lack

of facial expression datasets in real classroom settings. This dataset provides a reliable

foundation for future research and applications in educational technology, particularly in the

development of real-time emotion recognition models to enhance personalized learning and

teaching effectiveness.

Also, (W. Wang, 2024) stated that the problem of automatic detection in art teaching

classroom behavior, the research combines the YOLOv5 algorithm in the deep learning

algorithm and adds a two-way feature information pyramid function with weighting

capability to the neck part of the algorithm to achieve performance-based algorithm improve

ment. This research prunes and optimizes the model for the campus technology

implementation problem to improve the robustness and ease of implementation of the model.

The model is designed in line with the model of the art teaching classroom behavior training
set, and the applied experimental method is adopted for analysis. There fore, the research-

designed classroom behavior detection and analysis model for art teaching can effectively

detect the types of classroom behaviors of students in the process of art teaching with

excellent performance, providing an effective way to ensure the quality of student learning in

classroom teaching.

However, the education sector currently faces several challenges, including the subjectivity

of evaluation methods, uniformity of data, and a lack of real-time feedback. This study aims

to address these issues by leveraging deep learning techniques, specifically Convolutional

Neural Networks (CNNs), to accurately assess and enhance the quality of university teaching.

In contrast to traditional teaching quality assessment methods, which often lack rigor and

comprehensiveness, this study introduces a precise and thorough evaluation framework. By

integrating deep learning algorithms, the study seeks to improve the objectivity and accuracy

of evaluations, facilitate personalized feedback, and foster innovation in teaching

methodologies. This study not only addresses a gap in the field by utilizing multi-source data

for comprehensive evaluation but also validates the effectiveness of deep learning models in

assessing teaching quality. Additionally, the study provides a foundation for developing

targeted teaching improvement strategies(Gao, 2025).

Additionally, In the era of knowledge economy, social development has higher requirements

for artistic talents’ quality. The teaching mode of music and dance in colleges and

universities needs to be changed urgently. This study introduced deep learning theory and

designed a SPOC teaching mode for music and dance in universities to improve the quality of

music and dance teaching. Firstly, the K-means clustering algorithm was used to extract the

features of online learning behavior. Radial basis function neural network was used to predict

students’ performance to improve teachers’ mastery of students’ online learning behavior.

Subsequently, the teaching quality evaluation system was constructed using hierarchical
analysis. This indicates that the teaching mode gives full play to the advantages of SPOC

teaching, enhances students’ enthusiasm and initiative in learning, and improves their

professionalism and comprehensive quality(Ding, 2024).

As the researchers (Ren & Wu, 2023) said, the analysis of learning interactions during online

studying is a necessary task for designing online courses and sequencing key interactions,

which enables online learning platforms to provide users with more efficient and personalized

service. However, the research on predicting the interaction itself is not sufficient and the

temporal information of interaction sequences hasn’t been fully investigated. To fill in this

gap, based on the interaction data collected from Massive Open Online Courses (MOOCs),

this paper aims to simultaneously predict a user’s next interaction and the occurrence time to

that interaction. Three different neural network models: the long short-term memory, the

recurrent marked temporal point process, and the event recurrent point process, are applied

on the MOOC interaction dataset. It concludes that taking the correlation between the user

action and its occurrence time into consideration can greatly improve the model performance,

and that the prediction results are conducive to exploring dropout rates or online learning

habits and performances.

Meanwhile, with the advances in Artificial Intelligence (AI) and the increasing volume of

online educational data, Deep Learning techniques have played a critical role in predicting

student performance. Recent developments have assisted instructors in determining the

strengths and weaknesses of student achievement. This understanding will benefit from

adopting the necessary interventions to assist students in improving their performance,

helping at-risk of failure students, and preventing dropout rates. The educational prediction

findings hopefully serve as a strong foundation for administrators and instructors to observe

student performance and provide a suitable educational adaptation that can meet their needs

to protect them from failure and prevent their dropout(Alnasyan et al., 2024).
2.3 Student Performance

Student performance is the measurement of student achievement across various academic

subjects. Teachers and education officials typically measure achievement using classroom

performance, graduation rates and results from standardized tests (Konold et al., 2018). Also

Student performance is the extent to which a student, teacher or institution has attained their

short or long-term educational goals. Completion of educational benchmarks such as

secondary school diplomas and bachelor's degrees represent academic achievement (PB &

AV, 2020).

The Student performance is defined by students’ reporting of past semester CGPA/GPA and

their expected GPA for the current semester. The grade point average or GPA is now used by

most of the tertiary institutions as a convenient summary measure of the academic

performance of their students (Lwin, 2019). The GPA is a better measurement because it

provides a greater insight into the relative level of performance of individuals and different

group of students.

Academic achievement is commonly measured through examinations or continuous

assessments but there is no general agreement on how it is best evaluated or which aspects

are most important, procedural knowledge such as skills or declarative knowledge such

as facts. Furthermore, there are inconclusive results over which individual factors

successfully predict academic performance, elements such as test anxiety, environment,

motivation, and emotions require consideration when developing models of school

achievement (NUR NADIA et al., 2019).

2.4 Factors Influencing Academic Performance

2.4.1. Individual Differences Influencing Academic Performance

Individual differences in academic performance have been linked to differences

in intelligence and personality. Students with higher mental ability as demonstrated tend to
achieve highly in academic settings. A recent meta-analysis suggested that mental curiosity

(asmeasured by typical intellectual engagement) has an important influence on academic

achievement in addition to intelligence and conscientiousness (Kassarnig et al., 2018).

Children's semi-structured home learning environment transitions into a more structured

learning environment when children start first grade. Early academic achievement enhances

later academic achievement.

Parent's academic socialization is a term describing the way parents influence students'

academic achievement by shaping students' skills, behaviors and attitudes towards school.

Parents influence students through the environment and discourse parents have with their

children. Academic socialization can be influenced by parents' socio-economic status. Highly

educated parents tend to have more stimulating learning environments. Further, recent

research indicates that the relationship quality with parents will influence the development of

academic self-efficacy among adolescent-aged children, which will in turn affect their

academic performance (Kassarnig et al., 2018).

Studies have shown that physical activity can increase neural activity in the brain, specifically

increasing executive brain functions such as attention span and working memory and improve

academic performance in both elementary school children and college freshmen (Lu et al.,

2018).

2.4.2 Non-Cognitive Factors

Non-cognitive factors or skills, are a set of "attitudes, behaviors, and strategies" that promotes

academic and professional success, such as academic self-efficacy, self-control, motivation,

expectancy and goal setting theories, emotional intelligence, and determination.. The term

serves as a distinction of cognitive factors, which are measured by teachers through tests and

quizzes. Non-cognitive skills are increasingly gaining popularity because they provide a

better explanation for academic and professional outcomes (Tepper & Yourstone, 2018).
2.4.3 Self-efficacy

Self-efficacy is one of the best predictors of academic success. Self-efficacy is the belief of

being able to do something. (Hayat et al., 2020) looked at the Big Five traits on academic

success as well and saw that conscientiousness and emotional stability were predictors of

self-efficacy in over half of their analyses. However, self-efficacy was more indicative of

academic performance than personality in all of the analyses. This suggests that parents who

want their children to have academic achievement can look to increase their child's sense of

self-efficacy at school.

2.4.4 Motivation

Motivation is the reasoning behind an individual's actions. Research has found that students

with higher academic performance, motivation and persistence use intrinsic goals rather than

extrinsic ones (Bal-Ta\cstan et al., 2018). Furthermore, students who are motivated to

improve upon their previous or upcoming performance tend to perform better academically

than peers with lower motivation. In other words, students with higher need for achievement

have greater academic performance.

2.4.5 Self-control

Self-control, in the academic setting, is related self-discipline, self-regulation, delay of

gratification and impulse control. (Džinović et al., 2019) defined self-control as "the capacity

for altering one's own responses, especially to bring them into line with standards such as

ideals, values, morals, and social expectations, and to support the attainment of long-term

goals. In other words, self-control is the ability to prioritize long-term goals over the

temptation of short-term impulses.

High locus of control, where an individual attributes success to personal decision making and

positive behaviors such as discipline, is a ramification of self-control. High locus of control

has been found to have a positive predictive relationship with high collegiate GPA (Džinović

et al., 2019).

2.4.6 Extracurricular Activities

Organized extracurricular activities or cultural activities have yielded a positive relationship

with high academic performance including increasing attendance rates, school engagement,

GPA, postsecondary education, as well as a decrease in dropout rates and depression (Pinto &

He, 2019) . Additionally, positive developmental outcomes have been found in youth that

engage in organized extracurricular activities. High school athletics have been linked with

strong academic performance, particularly among urban youth. However, involvement in

athletics has been linked to increased alcohol consumption and abuse for high school students

along with increased truancy.

2.5 Machine Learning

Machine learning (ML) is a field of inquiry devoted to understanding and building methods

that 'learn', that is, methods that leverage data to improve performance on some set of tasks

(Mahesh, 2020). It is seen as a part of artificial intelligence. Machine learning algorithms

build a model based on sample data, known as training data, in order to make predictions or

decisions without being explicitly programmed to do so. Machine learning algorithms are

used in a wide variety of applications, such as in medicine, email filtering, speech

recognition, and computer vision, where it is difficult or unfeasible to develop conventional

algorithms to perform the needed tasks (Mahesh, 2020).

A subset of machine learning is closely related to computational statistics, which focuses on

making predictions using computers, but not all machine learning is statistical learning. The

study of mathematical optimization delivers methods, theory and application domains to the

field of machine learning. Data mining is a related field of study, focusing on exploratory

data analysis through unsupervised learning. Some implementations of machine learning use
data and neural networks in a way that mimics the working of a biological brain (Athey,

2018). In its application across business problems, machine learning is also referred to

as predictive analytics (Toğaçar et al., 2020).

2.5.1 Machine Learning Approaches

Machine learning approaches are traditionally divided into three broad categories, which

correspond to learning paradigms, depending on the nature of the "signal" or "feedback"

available to the learning system (Gupta & others, 2020).

2.5.2 Supervised Learning

Supervised learning algorithms build a mathematical model of a set of data that contains both

the inputs and the desired outputs. The data is known as training data, and consists of a set of

training examples. Each training example has one or more inputs and the desired output, also

known as a supervisory signal (Pant et al., 2020). In the mathematical model, each training

example is represented by an array or vector, sometimes called a feature vector, and the

training data is represented by a matrix (Rahman et al., 2020). Through iterative

optimization of an objective function, supervised learning algorithms learn a function that can

be used to predict the output associated with new inputs (Gabruseva et al., 2020). An optimal

function will allow the algorithm to correctly determine the output for inputs that were not a

part of the training data. An algorithm that improves the accuracy of its outputs or predictions

over time is said to have learned to perform that task (Hashmi et al., 2020). Types of

supervised-learning algorithms include active learning, classification and regression.

Classification algorithms are used when the outputs are restricted to a limited set of values,

and regression algorithms are used when the outputs may have any numerical value within a

range (Tang et al.,2022) . As an example, for a classification algorithm that filters emails, the
input would be an incoming email, and the output would be the name of the folder in which

to file the email.

2.5.3 Unsupervised Learning

Unsupervised learning algorithms take a set of data that contains only inputs, and find

structure in the data, like grouping or clustering of data points. The algorithms, therefore,

learn from test data that has not been labeled, classified or categorized. Instead of responding

to feedback, unsupervised learning algorithms identify commonalities in the data and react

based on the presence or absence of such commonalities in each new piece of data. A central

application of unsupervised learning is in the field of density estimation in statistics, such as

finding the probability density function. Though unsupervised learning encompasses other

domains involving summarizing and explaining data features (James et al., 2021)

Cluster analysis is the assignment of a set of observations into subsets (called clusters) so that

observations within the same cluster are similar according to one or more predesigned

criteria, while observations drawn from different clusters are dissimilar. Different clustering

techniques make different assumptions on the structure of the data, often defined by

some similarity metric and evaluated, for example, by internal compactness, or the similarity

between members of the same cluster, and separation, the difference between clusters. Other

methods are based on estimated density and graph connectivity (Ullah et al., 2021).

2.5.4 Semi-supervised learning

Semi-supervised learning falls between unsupervised learning (without any labeled training

data) and supervised learning (with completely labeled training data). Some of the training

examples are missing training labels, yet many machine-learning reseachers have found that

unlabeled data, when used in conjunction with a small amount of labeled data, can produce a

considerable improvement in learning accuracy (Zhou, 2021). In weakly supervised learning,

the training labels are noisy, limited, or imprecise; however, these labels are often cheaper to

obtain, resulting in larger effective training sets.

2.5.5 Reinforcement learning

Reinforcement learning is an area of machine learning concerned with how software

agents ought to take actions in an environment so as to maximize some notion of cumulative

reward. Due to its generality, the field is studied in many other disciplines, such as game

theory, control theory, operations project, information theory, simulation-based

optimization, multi-agent systems, swarm intelligence, statistics and genetic

algorithms(François-Lavet et al., 2018) . In machine learning, the environment is typically

represented as a Markov decision process (MDP). Many reinforcement learning algorithms

use dynamic programming techniques. Reinforcement learning algorithms do not assume

knowledge of an exact mathematical model of the MDP, and are used when exact models are

infeasible. Reinforcement learning algorithms are used in autonomous vehicles or in learning

to play a game against a human opponent (François-Lavet et al., 2018).

2.6 Deep Learning

Deep learning is an artificial intelligence (AI) technique that seeks to learn from experience to

resemble the human brain. Through a training procedure, these representations are learned.

To teach the software how to detect an object, we must first train it with a large number of

object images that we categorise according to different classes. Deep learning-based

algorithms, on average, require a large amount of training data and take longer to train than

traditional machine learning methods. Finding unique attributes when trying to recognise any

object or character on an image is time-consuming and complex. Unlike traditional machine

learning, where features are manually retrieved, problems can be solved using deep learning

approaches, which extract important characteristics from data automatically. A neural

network with multiple hidden layers is known as deep learning. They may build complicated
notions from simple concepts after an image has been taught over the network. By integrating

simple elements such as shape, edges, and corners, an image can be trained in the network to

learn items such as characters, faces, and so on. As the image travels through the layers, each

one gets a simple property while moving on to the next. As the layers grow larger, the

network may learn more complex features and eventually merge them to identify the image.

In the field of computer vision, deep learning has discovered a bunch of use. The domains

that work with facial data were among the most important computer vision applications.

2.6.1 How Deep Learning Works

Deep learning neural networks, or artificial neural networks, attempts to mimic the human

brain through a combination of data inputs, weights, and bias. These elements work together

to accurately recognize, classify, and describe objects within the data (Kelleher, 2019). Deep

neural networks consist of multiple layers of interconnected nodes, each building upon the

previous layer to refine and optimize the prediction or categorization. This progression of

computations through the network is called forward propagation. The input and output layers

of a deep neural network are called visible layers. The input layer is where the deep learning

model ingests the data for processing, and the output layer is where the final prediction or

classification is made.

Another process called backpropagation uses algorithms, like gradient descent, to calculate

errors in predictions and then adjusts the weights and biases of the function by moving

backwards through the layers in an effort to train the model. Together, forward propagation

and backpropagation allow a neural network to make predictions and correct for any errors

accordingly. Over time, the algorithm becomes gradually more accurate (Nikolenko et al.,

2018).
Most deep learning methods use neural network architectures, which is why deep learning

models are often referred to as deep neural networks. The term “deep” usually refers to the

number of hidden layers in the neural network. Traditional neural networks only contain 2-3

hidden layers, while deep networks can have as many as 150. Deep learning models are

trained by using large sets of labeled data and neural network architectures that learn features

directly from the data without the need for manual feature extraction (Kelleher, 2019).

2.6.2 MobileNetV2

MobileNetV2 is a convolutional neural network architecture that seeks to perform well on

mobile devices. It is based on an inverted residual structure where the residual connections

are between the bottleneck layers. The intermediate expansion layer uses lightweight

depthwise convolutions to filter features as a source of non-linearity (Altameem et al., 2021).

As a whole, the architecture of MobileNetV2 contains the initial fully convolution layer with

32 filters, followed by 19 residual bottleneck layers. In MobileNetV2, a better module is

introduced with inverted residual structure. Non-linearities in narrow layers are removed this

time (Jabbar et al., 2020). With MobileNetV2 as backbone for feature extraction, state-of-the-

art performances are also achieved for object detection and semantic segmentation as shown

in figure 1.
Figure 1: MobileNetV2 Architecture. Source: (Jabbar et al., 2020)

MobileNetV2 is a significant improvement over MobileNetV1 and pushes the state of the art

for mobile visual recognition including classification, object detection and semantic

segmentation (Gwak et al., 2020). MobileNetV2 is released as part of TensorFlow-Slim

Image Classification Library, or you can start exploring MobileNetV2 right away in

Colaboratory. MobileNetV2 builds upon the ideas from MobileNetV1, using depthwise

separable convolution as efficient building blocks. However, V2 introduces two new features

to the architecture: 1) linear bottlenecks between the layers, and 2) shortcut connections

between the bottlenecks1. The basic structure is shown in figure 2 below.

Figure 2: MobileNetV2 Architecture basic structure. Source: (Gwak et al., 2020)

The figure 2 above, state the overview of MobileNetV2 Architecture. Blue blocks represent

composite convolutional building blocks as shown above. The intuition is that the bottlenecks

encode the model’s intermediate inputs and outputs while the inner layer encapsulates the

model’s ability to transform from lower-level concepts such as pixels to higher level

descriptors such as image categories. Finally, as with traditional residual connections,

shortcuts enable faster training and better accuracy.

2.6.3 Convolutional Neural Network (CNN)

A CNN (convolutional neural network) is a kind of artificial neural network that is commonly

used for image or object identification and categorization. Using a CNN, Deep Learning

recognises items in an image. An input layer, hidden layers, and an output layer are all part of

a standard neural network. The anatomy of the brain inspired CNNs. Artificial neurons or

nodes in CNNs collect inputs, process them, and deliver the result as output, rather like a

neuron inside the brain functions and transmits signals between cells. The images are used as

a source of data. Multiple hidden layers may exist in CNNs, each of which performs feature

extraction from the image by performing calculations. The very first layer that extracts

feature out of an input image is convolution. The object is classified and identified in the

output layer by the fully connected layer. The convolutional layer is the most important

constituent of CNN. The mathematical procedure of convolution is used to combine two

sources of data. Gender estimation from social image collection, images that do not require

access to private details of the subject areas that are not displayed in the images, such as their

birth date, and the usual approach that includes the collection of other information about an

individual and on the basis about which we discover gender on manually handled annotated

data for gender recognition. That is why we use D-CNN, which works directly on images and

aids in precise gender estimation. Overfitting is usually.

In the field of artificial intelligence, CNNs are a kind of ‘Deep Neural Network’ (DNN) that

can identify and categorize certain aspects in pictures and are commonly used to

analyse visual data. Its uses span from video and picture identification to image

classification, image analysis in medicine, and computer vision and natural language

processing. When two functions are multiplied together, the result is a third function that

reflects how the form of one function is altered by the other. Multiplying two matrices
representable photos produces an output from which features from the image may be

extracted(Alqam et al., 2021).

Feature Extraction is a convolution technique that separates and identifies the image’s

distinct features for analysis. Convolution output is fed into a fully connected layer that uses

the information gathered in earlier rounds to forecast the image’s class. The CNN has three

layers: “convolutional, pooling, and fully-connected (FC) layers.” When these layers are put

together, a CNN will be formed.

Fig. CNN Architecture (Gaurav et al., 2023)

2.6.4 Convolution layer

The input pictures are initially processed via a convolutional layer, which is utilized to

extract different characteristics. The input picture is processed by this layer, which extracts

characteristics from it. This layer contains the picture matrix and the kernel/filter that extracts

the image's features. It is possible to get an image feature map of dimension (a-(fh+1), b-

(fw+1),1)) from an image matrix of volume (a). In this layer, the convolution of the input

picture with a filter of a certain size is carried out. Dot products are calculated by multiplying

the input image's pixels by the filter's

pixel size. We get information about the image's corners and edges from this layer's output,

which is called a Feature map. Afterwards, this ‘feature map’ is supplied to further layers in

order to explore more about the input picture.

Fig. convolutional layer (Gaurav et al., 2023)

2.6.5 Pooling layer

In order to decrease the algorithm's computing costs, this layer aims to reduce the size of the

'convolved feature map.' When the photos, like those we used in the project, are too huge,

the pooling layer decreases the number of parameters or blocks in the feature matrix.

Max Pooling: the filter area is pooled to its fullest extent.

Average Pooling: pools the average of the elements.

Sum Pooling: The total of the components in the filter is gathered in a pool called the sum

pooling.

The pooled feature map was created using Max Pooling for this project. The biggest element

on the feature map is used in Max Pooling.

Fig. Pooling layer (Gaurav et al., 2023)

2.7 Comparison of DT, KNN, and SVM.

Machine learning algorithms have become indispensable tools for analyzing and predicting

student performance. Among these, Decision Trees (DT), k-Nearest Neighbors (KNN), and

Support Vector Machines (SVM) stand out for their unique strengths and limitations. Each

algorithm offers different approaches to handling data, and their effectiveness varies

depending on the nature of the dataset and the specific application. Below is a comparison of

these three algorithms based on their functionality, performance, and suitability for

educational data analysis.

2.7.1 Decision Tree (DT)

Decision Trees are widely recognized for their simplicity and interpretability. They classify

data by creating a tree-like model of decisions based on the features of the dataset. The

algorithm splits the data iteratively at decision nodes using criteria such as Gini impurity or

information gain. DTs provide clear and easily interpretable rules, making them useful for

understanding the relationships between variables (Rastrollo-Guerrero et al., 2020).

2.7.2 k-Nearest Neighbors (KNN)

KNN is a simple, instance-based algorithm that classifies data by identifying the closest

training samples (neighbors) in the feature space (Cover & Hart, 1967). The algorithm

assigns a class based on the majority class among the k-nearest neighbors. KNN requires

storing the entire dataset and recalculating distances for every query, which makes it

computationally expensive for large datasets (Chitti et al., 2020).

2.7.3 Support Vector Machines (SVM)

SVM is a powerful supervised learning algorithm that finds the optimal hyperplane to

separate classes in the feature space. It uses a kernel trick to handle both linear and non-linear

data effectively. SVM requires careful tuning of parameters such as the kernel type,

regularization parameter (C), and gamma (for non-linear kernels), which can make it

challenging to optimize (Sathe & Adamuthe, 2021).

2.8 Review of Related Literature

However, (Sudais & Asad, 2022) ensure better comparison, by applying both Support Vector

Machine algorithm and K-Nearest Neighbor algorithm on the dataset to predict the student’s

grade and then compared their accuracy. Empirical studies outcome indicated that Support

Vector Machine achieved slightly better results with correlation coefficient of 0.96, while the

K-Nearest Neighbor achieved correlation coefficient of 0.95.

Furthermore, (Rastrollo-Guerrero et al., 2020) in their study, almost 70 papers were analyzed

to show different modern techniques widely applied for predicting students’ performance,

together with the objectives they must reach in this field. These techniques and methods,

which pertain to the area of Artificial Intelligence, are mainly Machine Learning,

Collaborative Filtering, Recommender Systems, and Artificial Neural Networks, among

others.

Also, (Chitti et al., 2020) in their study reviews EDM focusing on the factors influencing

student's predictions, various algorithms used, and identified the gaps. The study also gives

an insight into how the “black-box” decisions of the prediction model are made, the role of

various explainable AI (XAI) techniques in making the model results interpretable, and their

contribution to producing explainable results.

However, (Albreiki et al., 2021) in their systematic review, the relevant EDM literature

related to identifying student dropouts and students at risk from 2009 to 2021 is reviewed.

The review results indicated that various Machine Learning (ML) techniques are used to

understand and overcome the underlying challenges; predicting students at risk and students

drop out prediction. Moreover, most studies use two types of datasets: data from student

colleges/university databases and online learning platforms. ML methods were confirmed to

play essential roles in predicting students at risk and dropout rates, thus improving the

students’ performance.

Furthermore, (Ghorbani & Ghousi, 2020) proposed the Random hold-out and Shuffle 5-fold

cross-validation methods are used as model validation techniques. They achieved results

using different evaluation metrics indicate that fewer numbers of classes and nominal features

will lead models to better performance. Also, classifiers do not perform well with imbalanced

data, so solving this problem is necessary. The performance of classifiers is improved using

balanced datasets. Additionally, the results of the Friedman test, which is a statistical

significance test, confirm that the SVM-SMOTE is more efficient than the other resampling

methods. Moreover, The Random Forest classifier has achieved the best result among all

other models while using SVM-SMOTE as a resampling method.

In their paper, (Sathe & Adamuthe, 2021) presents investigation of application of C5.0, J48,

CART, Naïve Bayes (NB), K-Nearest Neighbour (KNN), Random Forest and Support Vector

Machine for prediction of students’ performance. Three datasets from school level, college

level and e-learning platform with varying input parameters are considered for comparison

between C5.0, NB, J48, Multilayer Perceptron (MLP), PART, Random Forest, BayesNet, and

Artificial Neural Network (ANN). Paper presents comparative results of C5.0, J48, CART,

NB, KNN, Random forest and SVM on changing tuning parameters. The performance of
these techniques is tested on three different datasets. Results show that the performances of

Random forest and C5.0 are better than J48, CART, NB, KNN, and SVM.

According to (Mengash, 2020), their study focuses on ways to support universities in

admissions decision making using data mining techniques to predict applicants’ academic

performance at university. A data set of 2,039 students enrolled in a Computer Science and

Information College of a Saudi public university from 2016 to 2019 was used to validate the

proposed methodology. The results demonstrate that applicants’ early university performance

can be predicted before admission based on certain pre-admission criteria (high school grade

average, Scholastic Achievement Admission Test score, and General Aptitude Test score).

The results also show that Scholastic Achievement Admission Test score is the pre-admission

criterion that most accurately predicts future student performance. Therefore, this score

should be assigned more weight in admissions systems. They also found that the Artificial

Neural Network technique has an accuracy rate above 79%, making it superior to other

classification techniques considered (Decision Trees, Support Vector Machines, and Naïve

Bayes).

In their study, (Hashim et al., 2020) compared the performances of several supervised

machine learning algorithms, such as Decision Tree, Naïve Bayes, Logistic Regression,

Support Vector Machine, K-Nearest Neighbour, Sequential Minimal Optimisation and Neural

Network. They trained a model by using datasets provided by courses in the bachelor study

programmes of the College of Computer Science and Information Technology, University of

Basra, for academic years 2017–2018 and 2018–2019 to predict student performance on final

examinations. Results indicated that logistic regression classifier is the most accurate in

predicting the exact final grades of students (68.7% for passed and 88.8% for failed).

According to (Sekeroglu et al., 2019) in their paper, two datasets have been considered for

the prediction and classification of student performance respectively using five machine
learning algorithms. Eighteen experiments have been performed and preliminary results

suggest that performances of students might be predictable and classification of these

performances can be increased by applying pre-processing to the raw data before

implementing machine learning algorithms.

Also, (Tomasevic et al., 2020) provide a comprehensive analysis and comparison of state of

the art supervised machine learning techniques applied for solving the task of student exam

performance prediction, i.e. discovering students at a “high risk” of dropping out from the

course, and predicting their future achievements, such as for instance, the final exam scores.

For both classification and regression tasks, the overall highest precision was obtained with

artificial neural networks by feeding the student engagement data and past performance data,

while the usage of demographic data did not show significant influence on the precision of

predictions. To exploit the full potential of the student exam performance prediction, it was

concluded that adequate data acquisition functionalities and the student interaction with the

learning environment is a prerequisite to ensure sufficient amount of data for analysis.

Furthermore, (X. Wang et al., 2020) stated that, due to the limited representation capability of

these manually extracted features, they can only understand the students’ behaviors

shallowly. To make the prediction process timely and automatically, they treat the

performance prediction task as a short-term sequence prediction problem, and propose a two-

stage classification framework, i.e., Sequence-based Performance Classifier (SPC), which

consists of a sequence encoder and a classic data mining classifier. More specifically, to

deeply discover the sequential features from students’ campus behaviors, they first introduce

an attention-based Hybrid Recurrent Neural Network (HRNN) to encode their recent

behaviors by giving a higher weight to the ones that are related to the students’ last action.

Then, to conduct student performance prediction, they further involve these learned features

to the classic Support Vector Machine (SVM) algorithm and finally achieve their SPC model.
They conduct extensive experiments in the real-world student card dataset. The experimental

results demonstrate the superiority of their proposed method in terms of Accuracy and Recall.

2.8 summary table authors, proposed topic, algorithm used, findings and limitations

Author(s) Proposed Algorithm/Technique Findings Limitations

Topic Used
Deng et al. Predicting Owl Search Optimized High accuracy Lack of
(2024) academic Dynamic Deep Neural and efficiency comparative
performance of Network (OSO- in predicting analysis with
primary school DDNN) academic alternative
learners based performance; models; results
on self-esteem OSO-DDNN specific to
and identifies at-risk dataset used
individuality students (LegiLexi
effectively. exam).
Al-azazi & Day-wise Artificial Neural ANN-LSTM Limited dataset
Ghurab multi-class Network and Long outperformed scope (specific
(2023) prediction of Short-Term Memory baseline models to MOOCs);
student (ANN-LSTM), RNN, with 70% 70% accuracy
performance in GRU accuracy, suggests
MOOCs showing potential for
improved early improvement.
predictions for
student
performance.
C. Wang Real-time Deep learning-based Established first Dataset limited
et al. emotion facial expression real-world to specific
(2024) recognition in recognition; OpenCV facial emotion
classrooms to for preprocessing expression categories;
enhance dataset with challenges in
personalized 5,527 images; broader
learning model enables classroom
real-time implementation.
emotion
recognition for
personalized
support in
classrooms.
W. Wang Automatic YOLOv5 algorithm Improved Focused only
(2024) detection of with a two-way accuracy on art teaching;
classroom feature information (0.973) after results may not
behavior in art pyramid function model generalize to
teaching optimization; other subjects
enhanced or
robustness and environments.
ease of
implementation
 for detecting
student
behaviors in art
classrooms.
Gao Improving Convolutional Neural High Multi-source
(2025) university Networks (CNNs) classification data integration
teaching accuracies for may require
quality using teaching quality high
deep learning ('Excellent': computational
92%, 'Good': resources;
88%, 'Average': limited to
85%, 'Poor': university-level
80%); objective settings.
evaluation
framework
developed.
Ding Enhancing K-means clustering, Improved Specific to
(2024) music and Radial Basis Function literacy scores music and
dance Neural Network and dance; limited
education (RBFNN), and music/dance scalability to
quality through hierarchical analysis scores (average other
SPOC teaching increase of 11 educational
model points); domains.
enhanced
student
engagement and
professionalism.
Ren & Wu Predicting Long Short-Term Improved Dataset limited
(2023) MOOC Memory (LSTM), prediction by to MOOCs;
interactions Recurrent Marked integrating user temporal
and occurrence Temporal Point action models may not
times Process, Event correlations generalize to
Recurrent Point with occurrence other
Process time; insights educational
into dropout platforms.
rates and
learning habits.
Alnasyan Review of deep Deep Neural Networks DNNs and Limited scope
et al. learning (DNNs), CNN-LSTM, CNN-LSTM of datasets
(2024) techniques for other DL techniques achieved high (mostly
predicting prediction MOOCs and
student accuracy LMSs); lack of
performance (>90%); focus on
learning explainability
behavior and of models.
activity features
are critical for
prediction.
Sudais & Comparison of Support Vector SVM performed Focused on
Asad SVM and KNN Machine (SVM) and slightly better binary
(2022) for predicting K-Nearest Neighbor (correlation classification;
student grades (KNN) coefficient: limited
0.96) than KNN exploration of
(0.95). other
algorithms or
hybrid
approaches.
Rastrollo- Modern Machine Learning, Highlights use Generalized
Guerrero techniques for Collaborative of AI findings; lacks
et al. predicting Filtering, techniques for detailed
(2020) student Recommender prediction; implementation
performance Systems, Artificial provides specifics for
Neural Networks comprehensive algorithms.
analysis of 70
papers.
Chitti et Explainable AI Explainable AI (XAI), Identified Lacks detailed
al. (2020) for student various ML algorithms factors evaluation
performance influencing metrics for
prediction predictions and compared
introduced XAI algorithms.
techniques to
make models
interpretable.
Albreiki et Systematic Various Machine Identified ML’s Limited
al. (2021) review of Learning (ML) role in exploration of
predicting techniques predicting at- DL techniques;
student risk students; results
dropouts datasets from dependent on
universities and dataset types
online platforms and
analyzed. preprocessing.
Ghorbani Enhancing ML Random Forest, SVM- SVM-SMOTE Results limited
& Ghousi model SMOTE, Random and Random to dataset
(2020) performance Hold-Out, Shuffle 5- Forest achieved properties; lack
for imbalanced Fold Cross-Validation the best of focus on
data performance; real-time
imbalanced data prediction.
handling critical
for accurate
predictions.
Sathe & Comparative Decision Trees, Random Forest Focused on
Adamuthe study of ML Random Forest, Naïve and C5.0 conventional
(2021) algorithms for Bayes, KNN, SVM, outperformed ML algorithms;
student ANN other models; limited
performance datasets from exploration of
prediction school, college, hybrid or deep
and e-learning learning
platforms approaches.
analyzed.
Mengash Supporting Artificial Neural ANN achieved Limited dataset
(2020) university Network (ANN), accuracy above (Computer
admissions Decision Trees, SVM,
79%; Scholastic Science
using data Naïve Bayes Achievement students only);
mining Admission Test results may not
identified as the generalize to
strongest pre- other fields.
admission
predictor.
Hashim et Supervised ML Decision Trees, Naïve Logistic Moderate
al. (2020) algorithms for Bayes, Logistic Regression was accuracy;
predicting Regression, SVM, most accurate limited focus on
student exam KNN, Neural Network for predicting feature
performance final grades selection and
(68.7% for imbalanced
passed, 88.8% data.
for failed).
Sekeroglu Prediction and Five ML algorithms Performance Focus on basic
et al. classification improved by ML techniques;
(2019) of student preprocessing lacks discussion
performance raw data; on DL models.
classification
accuracy
increased with
balanced
datasets.
Tomasevic Comprehensive Artificial Neural ANN achieved Limited
et al. analysis of ML Networks, student highest emphasis on
(2020) techniques for engagement and precision; explainability
predicting performance data demographic of predictions;
student exam data had relies heavily
performance minimal impact on student
on predictions. interaction data.
X. Wang Sequence- Hybrid Recurrent HRNN Dataset specific
et al. based student Neural Network effectively to student card
(2020) performance (HRNN), Support captured data; limited
prediction Vector Machine sequential generalizability
(SVM) features; to other
improved contexts.
accuracy and
recall for
student
performance
prediction.

2.9Research Gap
While numerous studies have made significant strides in leveraging machine learning (ML)

and deep learning (DL) techniques to predict student performance and improve educational

outcomes, several critical gaps remain unaddressed. Most existing studies focus on academic

and demographic features (e.g., past performance, learning behavior, attendance) for

prediction purposes (Rastrollo-Guerrero et al., 2020; Mengash, 2020). However,

psychological determinants like self-esteem and individuality, which are critical to academic

achievement, remain underexplored in prediction models. Only a few studies, such as Deng et

al. (2024), have incorporated these variables, suggesting the need for further exploration.

Many models, especially those applied to Massive Open Online Courses (MOOCs), provide

predictions only at the end of the course, delaying interventions (Al-azazi & Ghurab, 2023).

Real-time or day-wise prediction frameworks that enable in-time interventions are still

limited.

Research highlights the potential of facial expression recognition in understanding students'

emotional states for personalized learning (C. Wang et al., 2024). However, the availability of

real-world datasets for classroom-based emotion recognition is limited, restricting the

development of practical applications. Studies indicate that imbalanced datasets negatively

impact classifier performance (Ghorbani & Ghousi, 2020). Despite the availability of

advanced resampling methods like SVM-SMOTE, limited efforts have been made to

standardize their use across educational datasets. Current studies, such as Gao (2025), utilize

multi-source datasets (student feedback, course content, and teacher evaluations) but fail to

include diverse and comprehensive sources that encompass psychological, emotional, and

behavioral data in conjunction with academic performance. While models like YOLOv5

achieve high accuracy in classroom behavior detection for specific domains like art teaching

(W. Wang, 2024), their adaptability to other classroom settings remains unexplored.

Additionally, their reliance on pre-defined datasets restricts broader generalization.

Though explainable AI (XAI) approaches have been discussed to improve model

transparency and trustworthiness (Chitti et al., 2020), limited research integrates these

techniques into predictive models for student performance, leaving a gap in interpretability

for stakeholders like educators and parents. Many studies classify student performance into

binary categories, such as "pass" or "fail" (Albreiki et al., 2021; Sathe & Adamuthe, 2021).

The absence of multi-class classification limits nuanced insights into student progress.

Although DL models like ANN, LSTM, CNN, and HRNN have demonstrated high prediction

accuracy (Alnasyan et al., 2024; X. Wang et al., 2020), their combined potential for

integrating sequential, psychological, and emotional data remains unexplored.

 CHAPTER THREE: METHODOLOGY

3.1 Research Methodology

This is the method used in getting data for this research; the instruments used for collection of

data and the method of data analysis. It also gives an overview of the tools and techniques

employed for obtaining data. This formed the basis for the collection, analysis, interpretation,

and presentation of data for this project work. These involve: The method of study is based

on information from Internet surfing, searching for information from a library, and consulting

an expert on the topic concerned.

3.2 Model Adopted

The Agile SDLC model is a combination of iterative and incremental process models with a

focus on process adaptability and customer satisfaction through rapid delivery of working

software products. Agile methods break the product into small incremental builds. These

builds are provided in iterations. Each iteration typically lasts from about one to three weeks

or months. Every iteration involves cross-functional teams working simultaneously on

various areas like planning, requirements analysis, design, coding, unit testing, and

acceptance testing. At the end of the iteration, a working product is displayed to the customer

and important stakeholders.

The Agile model believes that every project needs to be handled differently and the existing

methods need to be tailored to best suit the project requirements. In agile, the tasks are

divided into time boxes or small-time frames to deliver specific features for a release. An

iterative approach is taken and working software is delivered after each iteration. Each build
is incremental in terms of features; the final build holds all the features required by the

customer. Here is a graphical illustration of the Agile Model.

Figure 1: Agile Model. Source: (Silva et al., 2020)

3.2.1 Following are the Agile Manifesto principles

 Individuals and interactions: In agile development, self-organization and

motivation are important, as are interactions like co-location and pair

programming.

 Working software: Working software is considered the best means of

communication with the customer to understand their requirements instead of just

depending on documentation.

 Customer collaboration: As the requirements cannot be gathered completely in

the beginning of the project due to various factors, continuous customer

interaction is very important to get proper product requirements.

  Responding to change: agile development is focused on quick responses to

change and continuous development.

3.3 Data Capture Method

During the gathering of data and facts used in this project to achieve a system, the research

made involved to investigate the extent to which counterfeiting has permeated the Nigeria

economy and propose a mobile based solution which consumers can use to effectively and

efficiently identify counterfeit products, by an expert, surfing of internet, searching of

information from library and also consulting of an expert on the topic concerned.

Although there are various methods of data collection, the researcher chose the two main

sources of data collection in carrying out the study. They are:

i. Secondary source

ii. Primary source

3.3.1 Primary Source

Primary Source refers to the sources of collecting original data in which the researcher made

use of empirical approach i.e. interview. Furthermore, the researcher adopted oral

interviewing: The interview method of data collection can be defined as a systematic way of

collecting data or information from a respondent through asking questions directly from the

respondent and also collecting information to facilitate understanding. The oral interview was

conducted between the researcher and some resource persons. Reliable facts were gotten

based on the questions posed to the research persons which help the work in the area of

solution presentation of the new design. The researcher adopts the method of unstructured

interviews.

3.3.2 Secondary Sources

Secondary Sources data from this kind of project cannot be overemphasized. The secondary

data was obtained from magazines, journals, newspapers and library sources. Manuals and
reports based on feedback were obtained and studied and a lot of information concerning the

system to be produced was obtained. However, secondary source data was mostly adopted in

this research, we were able to get secondary resources from various academic databases such

as Elsevier, Google Scholar, Research Gate, Science Direct, and IEEE etc.

3.4 System Analysis

System analysis is the process of collecting and analyzing facts. A comprehensive analysis

was carried out in order to define the requirements for designing the new system. The new

system will rectify the problems of the old system.

3.4.1 Analysis of the Existing System

In today's world, Prediction of student’s performance became an urgent desire in most of

educational entities and institutes. That is essential in order to help at-risk students and assure

their retention, providing the excellent learning resources and experience, and improving the

institutions ranking and reputation (Agrawal & Mavani, 2019).

These days the lack of existing system to analyses and judge the student’s performance and

progress is not being addressed (Tomasevic et al., 2020). There are two reasons why this is

often happening. First, the present system is not accurate to predict students‟ performance.

Second, because of shortage of consideration of some vital factor that are affecting students‟

performance. Forecasting student performance is essential for educators to obtain early

feedback and take immediate action or early precautions if necessary to improve the student’s

performance (Mengash, 2020). This prediction can be managed by locating the source of the

problem. Should it be from extra activities that the student is participating in, family

problems, or health problems? All these factors can have a major effect on student

performance (Sathe & Adamuthe, 2021). There is often a great need to be able to predict
future students’ behavior in order to improve curriculum design and plan interventions for

academic support and guidance on the curriculum offered to the students.

However, there is no artificial intelligence approach in predicted of student performance in

the Department of Computer Science Federal Polytechnic Bauchi, most time manually

method of evaluating student performance is been deployed which lead to the lack of

efficiency and accuracy due to the highly competitive and complex environment. Thus,

analyzing performance, providing high-quality education, formulating strategies for

evaluating the students’ performance, and identifying future needs are some challenges faced

by the institutions today.

Student performance prediction at entry-level and during the subsequent periods will helps

the institutions effectively develop and evolve the intervention plans, where both the

management and educators are the beneficiaries of the students’ performance prediction

plans. However, this research work proposed a comparison evaluation of the following Deep

learning algorithms;Decision Tree, KNN and Support Vector Machines for student

performance predictions. In other, to know the best fit model among the three algorithms.

3.4.2 Problem of the Existing System

From the systems analysis above, the following observations were made about the existing

system:

 The present system is not accurate to predict students‟ performance

 Shortage of consideration of some vital factor that are affecting students’

performance.

 The lack of existing system to analyses and judge the student’s performance and

progress is not being addressed

3.4.3 Analysis of the Propose System

The prediction analysis is the approach which predicts future possibilities from the previous

data. The student performance prediction technique has the three phases which are pre-

processing, feature extraction and classification. This research work propose the development

of models using the following Deep learning algorithms; Decision Tree, k-nearest neighbor,

and support vector machine for student performance prediction using a dataset provided by

the University of Minho in Portugal, and it consists of 395 data samples. To ensure better

comparison, we applied both Support Vector Machine algorithm, logistic regression and K-

Nearest Neighbor algorithm on the dataset to predict the student’s performance and then

compared their accuracy to determine the best fit model

3.4.4 Feature of the Propose System

i. To perform data processing

a. Sometimes, datasets came up with non-numerical values and it is impossible to

give it to any classifier. The non-numeric dataset are convert to numerical ones.

b. Feature scaling is a method used to normalize the range of independent variables

or features of data. In data processing, it is also known as data normalization and

is generally performed during the data preprocessing step.

ii. To developed and train models using the following supervised machine learning

algorithms; Decision Tree, K-nearest neighbor, and support vector machine.

a. To use the developed model to predict student performance.

b. To compared the developed models and determine the best fit model among them.

3.5 System Design

As system analysis describes what a system should do to meet the information needs of users,

system design specifies how the system will accomplish the result of the analysis. System
design consists of design activities that produce system specifications that satisfy the

functional requirements developed in the system analysis.

3.5.1 Proposed System Framework

Dataset

Pre-Process

Training 70% Testing 30%

Train Models

Perform Predictions

Compare Model

Display Best fit

Model

Figure 2: Proposed system framework

The figure 3 above, illustrates the proposed system framework

3.5.2 Dataset Description

The dataset in this research work was provided by the University of Minho in Portugal, and it

consists of 395 data samples with the following attribute description; the dataset includes 33

attributes. Four of the attributes are nominal, 13 of the attributes are binary and 16 of the

attributes are numeric. The dataset includes 395 instances with no missing values.

3.5.3 Data processing

Before working with any dataset, there is a need to process the dataset so that, it will be ready

for training the models. Sometimes, datasets came up with non-numerical values and it is

impossible to give it to any classifier. The non-numeric values in dataset is converted to

numerical ones. Feature scaling was perform to normalize the range of independent variables

or features of data. In data processing, it is also known as data normalization and is generally

performed during the data preprocessing step. This will help our learning algorithms to

converge quickly.

3.5.4 Data Visualisation

Firstly we look deeper into some features by using multiple methods of visualisation such as

distribution plot, and Density etc. In other to have a good understanding of which features are

most impactful for student's performances.

3.5.5 Model implementation

The dataset will be divided into two 70% for the training and 30% for the testing. Hence, the

three models will be developed to predict student performance base on the three supervised

machine learning classifiers (Decision Tree, K-nearest neighbor and support vectors

machine).

3.5.6 Model Evaluation

In order to evaluate our models, we will calculate the accuracy of the model, visualize the

confusion matrix, and then plot the ROC curve. We will get two values of the accuracy, one

obtained with the training set and other with the test set. The test set accuracy is more

relevant for evaluating the performance on unseen data since it’s not biased. The developed

models will be evaluated based on the following evaluation metrics;

 F1 score.

 Accuracy score.

 Confusion matrix.
  ROC curve.

 ROC score.

3.5.7 Comparison of the three algorithms

The three classifiers that have been train to predict student performance will be compare

based on the evaluation metrics as stated above in other to determine the best fit model

among the three classifiers.

3.6 System Block Diagram

Figure 4 depicts the system block diagram, which shows the interaction of all the modules

that make up the entire proposed system.

Dataset Preprocessing Feature Scaling

Train_Test_Split

Perform
Train Models
Classification

Compared Models

Figure 3: System Block Diagram

3.7 ALGORITHM
Start

Dataset

Preprocess

& Feature Scale

Divide dataset into

Train and Test

Developed Models

No

Can Models predict

well?

Yes

Perform Prediction

Compared Models

Best Model

End

Figure 4: System Algorithm.

The system flowchart, as shown above, illustrates the sequential and logical flow of the

proposed system.

3.8 Adopted Programming Language

Python is a multi-paradigm programming language. Object-oriented programming and

structured programming are fully supported, and many of their features support functional
programming and aspect-oriented programming (including metaprogramming and

metaobjects) (Python, 2021). Many other paradigms are supported via extensions, including

design by contract and logic programming. Python uses dynamic typing and a combination of

reference counting and a cycle-detecting garbage collector for memory management. It uses

dynamic name resolution (late binding), which binds method and variable names during

program execution.

Its design offers some support for functional programming in the Lisp tradition. It has filter,

map and reduce functions; list comprehensions, dictionaries, sets, and generator expressions.

The standard library has two modules (itertools and functools) that implement functional

tools borrowed from Haskell and Standard ML (Fontenrose, 2022). Its core philosophy is

summarized in the document “The Zen of Python (PEP 20)”, which includes aphorisms such

as:

i. Beautiful is better than ugly.

ii. Explicit is better than implicit.

iii. Simple is better than complex.

iv. Complex is better than complicated.

v. Readability counts.

Rather than building all of its functionality into its core, Python was designed to be highly

extensible via modules. This compact modularity has made it particularly popular as a means

of adding programmable interfaces to existing applications (Martelli et al., 2023). Van

Rossum's vision of a small core language with a large standard library and easily extensible

interpreter stemmed from his frustrations with ABC, which espoused the opposite approach.

3.9 System Requirement

System requirement for the use of this research work includes the hardware and software

requirement specification.
3.9.1 Recommendation Requirement

The implementation was conducted on Python 3.8, tensorflow, Google Colab, and keras etc.

Running a Microsoft Window 11 PRO with the following specification:

i System Type: x64-based processor, 64-bit Operating System

ii Memory installed on system (RAM): 16.00 GB

iii Processor: Intel(R) Core (TM) i3-4000M @ 2.4 GHz

References

Al-azazi, F. A., & Ghurab, M. (2023). ANN-LSTM: A deep learning model for early student

performance prediction in MOOC. Heliyon, 9(4), e15382.

https://doi.org/10.1016/j.heliyon.2023.e15382
Albreiki, B., Zaki, N., & Alashwal, H. (2021). A systematic literature review of

student’performance prediction using machine learning techniques. Education Sciences,

11(9), 552.

Alnasyan, B., Basheri, M., & Alassafi, M. (2024). The power of Deep Learning techniques

for predicting student performance in Virtual Learning Environments: A systematic

literature review. Computers and Education: Artificial Intelligence, 6(December 2023),

100231. https://doi.org/10.1016/j.caeai.2024.100231

Alqam, M. H., Abu-Khamsin, S. A., Sultan, A. S., Al-Afnan, S. F., & Alawani, N. A. (2021).

An investigation of factors influencing carbonate rock wettability. Energy Reports, 7,

1125–1132. https://doi.org/10.1016/j.egyr.2021.01.091

Altameem, A., Kumar, A., Poonia, R. C., Kumar, S., & Saudagar, A. K. J. (2021). Early

identification and detection of driver drowsiness by hybrid machine learning. IEEE

Access, 9, 162805–162819.

Athey, S. (2018). The impact of machine learning on economics. In The economics of

artificial intelligence: An agenda (pp. 507–547). University of Chicago Press.

Bal-Ta\cstan, S., Davoudi, S. M. M., Masalimova, A. R., Bersanov, A. S., Kurbanov, R. A.,

Boiarchuk, A. V, & Pavlushin, A. A. (2018). The impacts of teacher’s efficacy and

motivation on student’s academic achievement in science education among secondary

and high school students. EURASIA Journal of Mathematics, Science and Technology

Education, 14(6), 2353–2366.

Chitti, M., Chitti, P., & Jayabalan, M. (2020). Need for interpretable student performance

prediction. 2020 13th International Conference on Developments in ESystems

Engineering (DeSE), 269–272.

Deng, J., Huang, X., & Ren, X. (2024). A multidimensional analysis of self-esteem and

individualism: A deep learning-based model for predicting elementary school students’

academic performance. Measurement: Sensors, 33(November 2023), 101147.

https://doi.org/10.1016/j.measen.2024.101147

Ding, J. (2024). Deep learning perspective on the construction of SPOC teaching model of

music and dance in colleges and universities. Systems and Soft Computing, 6(January),

200137. https://doi.org/10.1016/j.sasc.2024.200137

Džinović, V., \DJević, R., & \DJerić, I. (2019). The role of self-control, self-efficacy,

metacognition, and motivation in predicting school achievement. Psihologija, 52(1), 35–

52.

François-Lavet, V., Henderson, P., Islam, R., Bellemare, M. G., Pineau, J., & others. (2018).

An introduction to deep reinforcement learning. Foundations and Trends®in Machine

Learning, 11(3–4), 219–354.

Gabruseva, T., Poplavskiy, D., & Kalinin, A. (2020). Deep learning for automatic pneumonia

detection. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern

Recognition Workshops, 350–351.

Gao, Y. (2025). Deep learning-based strategies for evaluating and enhancing university

teaching quality. Computers and Education: Artificial Intelligence, 8(September 2024),

100362. https://doi.org/10.1016/j.caeai.2025.100362

Gaurav, A., Gupta, B. B., Chui, K. T., Arya, V., & Chaurasia, P. (2023). Deep Learning

Based Hate Speech Detection on Twitter. IEEE International Conference on Consumer

Electronics - Berlin, ICCE-Berlin, 2023. https://doi.org/10.1109/ICCE-

Berlin58801.2023.10375620
Ghorbani, R., & Ghousi, R. (2020). Comparing different resampling methods in predicting

students’ performance using machine learning techniques. IEEE Access, 8, 67899–

67911.

Gupta, R., & others. (2020). A survey on machine learning approaches and its techniques.

2020 IEEE International Students’ Conference on Electrical, Electronics and Computer

Science (SCEECS), 1–6.

Gwak, J., Hirao, A., & Shino, M. (2020). An investigation of early detection of driver

drowsiness using ensemble machine learning based on hybrid sensing. Applied Sciences,

10(8), 2890.

Hasan, H. M. R., Rabby, A. K. M. S. A., Islam, M. T., & Hossain, S. A. (2019). Machine

learning algorithm for student’s performance prediction. 2019 10th International

Conference on Computing, Communication and Networking Technologies (ICCCNT),

1–7.

Hashim, A. S., Awadh, W. A., & Hamoud, A. K. (2020). Student performance prediction

model based on supervised machine learning algorithms. IOP Conference Series:

Materials Science and Engineering, 928(3), 32019.

Hashmi, M. F., Katiyar, S., Keskar, A. G., Bokde, N. D., & Geem, Z. W. (2020). Efficient

pneumonia detection in chest xray images using deep transfer learning. Diagnostics,

10(6), 417.

Hayat, A. A., Shateri, K., Amini, M., & Shokrpour, N. (2020). Relationships between

academic self-efficacy, learning-related emotions, and metacognitive learning strategies

with academic performance in medical students: a structural equation model. BMC

Medical Education, 20(1), 1–11.

Jabbar, R., Shinoy, M., Kharbeche, M., Al-Khalifa, K., Krichen, M., & Barkaoui, K. (2020).

Driver drowsiness detection model using convolutional neural networks techniques for

android application. 2020 IEEE International Conference on Informatics, IoT, and

Enabling Technologies (ICIoT), 237–242.

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2021). Unsupervised learning. In An

introduction to statistical learning (pp. 497–552). Springer.

Kassarnig, V., Mones, E., Bjerre-Nielsen, A., Sapiezynski, P., Dreyer Lassen, D., &

Lehmann, S. (2018). Academic performance and behavioral patterns. EPJ Data Science,

7, 1–16.

Kelleher, J. D. (2019). Deep learning. MIT press.

Konold, T., Cornell, D., Jia, Y., & Malone, M. (2018). School climate, student engagement,

and academic achievement: A latent variable, multilevel multi-informant examination.

Aera Open, 4(4), 2332858418815661.

Laakel Hemdanou, A., Lamarti Sefian, M., Achtoun, Y., & Tahiri, I. (2024). Comparative

analysis of feature selection and extraction methods for student performance prediction

across different machine learning models. Computers and Education: Artificial

Intelligence, 7(October), 100301. https://doi.org/10.1016/j.caeai.2024.100301

Lu, O. H. T., Huang, A. Y. Q., Huang, J. C. H., Lin, A. J. Q., Ogata, H., & Yang, S. J. H.

(2018). Applying learning analytics for the early prediction of Students’ academic

performance in blended learning. Journal of Educational Technology \& Society, 21(2),

220–232.

Lwin, C. W. (2019). THE EFFECT OF SOCIAL MEDIA ON STUDENTS’ACADEMIC

PERFORMANCE (A CASE STUDY OF YANGON UNIVERSITY OF ECONOMICS).

 MERAL Portal.

Mahesh, B. (2020). Machine learning algorithms-a review. International Journal of Science

and Research (IJSR).[Internet], 9, 381–386.

Mengash, H. A. (2020). Using data mining techniques to predict student performance to

support decision making in university admission systems. IEEE Access, 8, 55462–

55470.

Nikolenko, S., Kadurin, A., & Arkhangelskaya, E. (2018). Deep learning. SPb.: Peter.

NUR NADIA, R., AHMAD, N., NUR FAIZA, M. A., & MUHAMAD SHAMSUL, I.

(2019). THE RELATIONSHIP BETWEEN THE IMPACT OF SOCIAL MEDIA AND

STUDENTS’ACADEMIC PERFORMANCE. The Asian Journal of Professional and

Business Studies, 2(2).

Pant, A., Jain, A., Nayak, K. C., Gandhi, D., & Prasad, B. G. (2020). Pneumonia detection:

An efficient approach using deep learning. 2020 11th International Conference on

Computing, Communication and Networking Technologies (ICCCNT), 1–6.

PB, B., & AV, B. (2020). A Study on Academic Achievement of Residential and Non-

residential secondary school students in relation to Locality. International Research

Journal on Advanced Science Hub, 2, 73–77.

Pinto, L. H., & He, K. (2019). ‘In the eyes of the beholder’: the influence of academic

performance and extracurricular activities on the perceived employability of Chinese

business graduates. Asia Pacific Journal of Human Resources, 57(4), 503–527.

Rahman, T., Chowdhury, M. E. H., Khandakar, A., Islam, K. R., Islam, K. F., Mahbub, Z. B.,

Kadir, M. A., & Kashem, S. (2020). Transfer learning with deep convolutional neural

network (CNN) for pneumonia detection using chest X-ray. Applied Sciences, 10(9),
 3233.

Rastrollo-Guerrero, J. L., Gómez-Pulido, J. A., & Durán-Dom\’\inguez, A. (2020). Analyzing

and predicting students’ performance by means of machine learning: A review. Applied

Sciences, 10(3), 1042.

Ren, J., & Wu, S. (2023). Prediction of user temporal interactions with online course

platforms using deep learning algorithms. Computers and Education: Artificial

Intelligence, 4(March), 100133. https://doi.org/10.1016/j.caeai.2023.100133

Sathe, M. T., & Adamuthe, A. C. (2021). Comparative Study of Supervised Algorithms for

Prediction of Students’ Performance. International Journal of Modern Education \&

Computer Science, 13(1).

Sekeroglu, B., Dimililer, K., & Tuncal, K. (2019). Student performance prediction and

classification using machine learning algorithms. Proceedings of the 2019 8th

International Conference on Educational and Information Technology, 7–11.

Sudais, M., & Asad, D. (2022). Student’s Academic Performance Prediction--A Review.

Tang, P., Yang, P., Nie, D., Wu, X., Zhou, J., & Wang, Y. (2022). Unified medical image

segmentation by learning from uncertainty in an end-to-end manner. Knowledge-Based

Systems, 241, 108215.

Tepper, R. J., & Yourstone, S. A. (2018). Beyond ACT \& GPA: self-efficacy as a non-

cognitive predictor of academic success. International Journal of Accounting \&

Information Management.

Toğaçar, M., Ergen, B., Cömert, Z., & Özyurt, F. (2020). A deep feature learning model for

pneumonia detection applying a combination of mRMR feature selection and machine

learning models. Irbm, 41(4), 212–222.

Tomasevic, N., Gvozdenovic, N., & Vranes, S. (2020). An overview and comparison of

supervised data mining techniques for student exam performance prediction. Computers

\& Education, 143, 103676.

Ullah, M. S., Qayoom, H., & Hassan, F. (2021). Viral Pneumonia Detection Using Modified

GoogleNet Through Lung X-rays. 2021 4th International Symposium on Advanced

Electrical and Communication Technologies (ISAECT), 1–6.

Wang, C., Wang, H., Hu, Z., & Chen, X. (2024). Annotated emotional image datasets of

Chinese university students in real classrooms for deep learning. Data in Brief, 57,

111147. https://doi.org/10.1016/j.dib.2024.111147

Wang, W. (2024). Application of deep learning algorithm in detecting and analyzing

classroom behavior of art teaching. Systems and Soft Computing, 6(January), 200082.

https://doi.org/10.1016/j.sasc.2024.200082

Wang, X., Yu, X., Guo, L., Liu, F., & Xu, L. (2020). Student performance prediction with

short-term sequential campus behaviors. Information, 11(4), 201.

Zhou, Z.-H. (2021). Semi-supervised learning. In Machine Learning (pp. 315–341). Springer.