The current issue and full text archive of this journal is available on Emerald Insight at:

https://www.emerald.com/insight/2046-3162.htm

Asian Education
Application of machine learning in and Development
Studies
higher education to predict students’
performance, learning engagement
and self-efficacy: a systematic
literature review Received 2 August 2024
Revised 5 December 2024
Accepted 7 January 2025
Juntao Chen
Faculty of Applied Sciences, Macao Polytechnic University, Macao, China and
Hainan College of Economics and Business, Haikou, China, and
Xiaodeng Zhou, Jiahua Yao and Su-Kit Tang
Faculty of Applied Sciences, Macao Polytechnic University, Macao, China

Abstract
Purpose – In recent years, studies have shown that machine learning significantly improves student performance
and retention and reduces the risk of student dropout and withdrawal. However, there is a lack of empirical research
reviews focusing on the application of machine learning to predict student performance in terms of learning
engagement and self-efficacy and exploring their relationships. Hence, this paper conducts a systematic research
review on the application of machine learning in higher education from an empirical research perspective.
Design/methodology/approach – This systematic review examines the application of machine learning (ML)
in higher education, focusing on predicting student performance, engagement and self-efficacy. The review
covers empirical studies from 2016 to 2024, utilizing a PRISMA framework to select 67 relevant articles from
major databases.
Findings – The findings show that ML applications are widely researched and published in high-impact
journals. The primary functions of ML in these studies include performance prediction, engagement analysis
and self-efficacy assessment, employing various ML algorithms such as decision trees, random forests, support
vector machines and neural networks. Ensemble learning algorithms generally outperform single algorithms
regarding accuracy and other evaluation metrics. Common model evaluation metrics include accuracy, F1 score,
recall and precision, with newer methods also being explored.
Research limitations/implications – First, empirical research literature was selected from only four renowned
electronic journal databases, and the literature was limited to journal articles, with the latest review literature and
conference papers published in the form of conference papers also excluded, which led to empirical research not
obtaining the latest views of researchers in interdisciplinary fields. Second, this review focused mainly on the
analysis of student grade prediction, learning engagement and self-efficacy and did not study students’ risk, dropout
rates, retention rates or learning behaviors, which limited the scope of the literature review and the application field
of machine learning algorithms. Finally, this article only conducted a systematic review of the application of
machine learning algorithms in higher education and did not establish a metadata list or carry out metadata analysis.
Originality/value – The review highlights ML’s potential to enhance personalized education, early intervention
and identifying at-risk students. Future research should improve prediction accuracy, explore new algorithms
and address current study limitations, particularly the narrow focus on specific outcomes and lack of
interdisciplinary perspectives.
Keywords Machine learning, Students’ performance, Self-efficacy, Engagement, Systematic literature review,
Empirical research
Paper type Literature review

Funding: This work was also supported by a research grant (No. RP/FCA-09/2023), offered by Macao
Polytechnic University, the Hainan Provincial Natural Science Foundation of China (project number:
622RC723), and the Research and Innovation Team for the Application of HyperAutomation Technology
(project number: HNJMT2023-101).
Availability of data and materials: The authors confirm that the data supporting the findings of this Asian Education and Development
Studies
study are available from the corresponding author upon reasonable request. © Emerald Publishing Limited
Conflict of interest: The authors declare that the research was conducted in the absence of any e-ISSN: 2046-3170
p-ISSN: 2046-3162
commercial or financial relationships that could be construed as potential conflicts of interest. DOI 10.1108/AEDS-08-2024-0166
AEDS 1. Introduction
With the wide application of educational big data in higher education, Educational Data
Mining (EDM) has gradually become a hot topic of concern for researchers, especially in
recent years, and the application of machine learning in education has made learning
prediction an important part of learning analytics. EDM uses various data mining tools and
methods to examine information in the academic environment, providing a deep
understanding of teaching and learning activities. This enables more timely educational
management, student performance analysis, and a better understanding of the learning process
(Hooda et al., 2022). EDM is the application of data mining methods to educational data, and
the five main functions of educational data mining include prediction, analysis, clustering,
relationship mining, and modeling studies (Baker and Yacef, 2009). A report by Nti et al.
(2022) shows that using machine learning algorithms to predict the impact of social media on
students’ academic performance, pointing out that the use of social media in the classroom
significantly negatively impacts students’ GPAs. Sugden et al. (2021) reported that the
K-Medoids algorithm was used to perform cluster analysis on students’ online learning
activities. They found that online activities contribute to deep learning and participation.
Learning prediction research refers mainly to predicting students’ future outcomes, such as
academic performance, learning goals, and learning abilities, based on prior and current
learning activities or learning outcomes, and its main role is to discover meaningful patterns
and hidden information, knowledge, and regularity from a large amount of learning activity
data (Meghji et al., 2023).
Moreover, due to the digitalization of the educational process, universities have
accumulated a large amount of electronic data about students and mined these data to
enable a better understanding and mapping of students’ learning behavior, the current status of
their learning, and other aspects of their learning, which is particularly important (Asif et al.,
2017). In the context of big data analytics, machine learning applications (e.g. constructing
predictive models) have been used to predict students’ course grades, final grades, dropout
rates, and academic risk (Wang, 2022; Aydo� gdu, 2020; Sultana et al., 2017; Veerasamy et al.,
2022) and to enhance students’ learning engagement, etc. (Moubayed et al., 2021; Wan et al.,
2019). Various machine learning algorithms (e.g. decision trees (DTs), support vector
machines (SVMs), gradient boosting (GB), neural networks (NNs), K-nearest neighbors
(KNNs), naı€ve Bayes (NB), and random forests (RFs)) have been widely used to create models
for learning behavior prediction, motivation, self-efficacy (De Backer et al., 2022a, b; Riestra-
Gonz�alez et al., 2021) and learning log analysis (Abdullah et al., 2023; Hang et al., 2022;
Huang et al., 2023).
Overall, as education systems focus on successful student completion and school
reputation, machine learning algorithms, which provide powerful tools and technical support
to improve the quality of higher education, have become particularly important in predicting
student academic performance and the early identification of at-risk students. Previous
literature reviews have provided insights in the areas of reviewing machine learning for
predicting academic performance and academic risk. For example, Luan and Tsai (2021)
reviewed empirical research on using machine learning to achieve precision education and
found that most studies focus on predicting learning performance or dropout risk. However,
there is a lack of integration of learning theories, and the research scope is mainly limited to
specific disciplines and educational levels. Two reports by Andrade-Gir� on et al. (2023) and
Fahd et al. (2022) showed that through evaluating the success rate of machine learning
algorithms in predicting student dropout, they found that the random forest algorithm
performed the best in predicting student dropout, with an accuracy of 99%. However the
research results lacked robustness and universality, and the model’s generalization ability was
limited. Kaddoura et al. (2022) found that the random forest algorithm performed the best in
predicting student dropout, with an accuracy of 99%; However, there is a lack of in-depth
exploration of the interpretability, privacy protection, and security of machine learning
models. Pelima et al. (2024) reviewed the application of machine learning in predicting
student graduation in higher education and found that support vector machines and random Asian Education
forests algorithms were the most commonly used algorithms, but research is insufficient, and Development
including issues with data accuracy and reliability. Okewu et al. (2021) found through research Studies
on the application of artificial intelligence neural networks in higher education data mining
that hardware challenges, training challenges, theoretical challenges, and quality issues were
the main obstacles to the application of neural networks in higher education. The limitation of
the literature was that it did not cover research from non-English sources. Umer et al. (2023)
thought that although there were multiple methods for predicting students’ academic
performance, there were still inconsistencies in the dataset, diversity in sample size, and
feature selection in the study. (Ili�c et al., 2021; Pelima et al., 2024) explored the current status
and potential of the application of artificial intelligence and machine learning technology in
higher education in Serbia and found that machine learning technology can help improve
students’ academic skills and participation. However, the research is limited by the small
sample size and does not cover all factors that may affect the application of technology.
While there is a growing body of literature on the application of machine learning for
predicting and analyzing academic performance, self-efficacy, and learning engagement in the
context of the big data era, there is a significant research gap. Specifically, there is a lack of
systematic reviews exploring the empirical effects of machine learning applications in
academic performance prediction. This study aims to fill this gap by systematically reviewing
empirical studies of machine learning in higher education, focusing on the purpose, algorithms
used, and the effectiveness of predictions in applications such as predicting students’ academic
performance, learning engagement, and self-efficacy in higher education. Our review differs
from existing literature by providing a comprehensive analysis of the current state of machine
learning applications, offering a comparative study of various algorithms used, and
highlighting the potential for enhancing personalized education, early intervention, and
identifying at-risk students. This systematic review will contribute to the body of knowledge
by identifying areas for future research and suggesting ways to improve prediction accuracy
and algorithmic robustness.

2. Literature review
2.1 Systematic overview of student performance prediction
Machine learning techniques have been used to predict students’ academic performance,
particularly to discover features that help students understand their successes and failures and
implement appropriate strategies to improve their performance (Qu et al., 2018). For example,
Huang et al. (2023) develop a personalized intervention tutoring learning approach that
combines quantitative analysis and qualitative feedback, uses machine learning algorithms to
construct a predictive model, performs feature selection to improve predictive performance,
and investigates the potential of self-regulated learning (SRL) to improve students’ academic
performance. Tseng et al. (2023) constructed models based on data such as personality traits
and assessed their potential to improve academic performance. Silva et al. (2024) analyzed
how students’ extracurricular activities and daily behaviors affect final academic performance
through a supervised machine learning approach. Nti et al. (2022) explored the effects of
nature, frequency, and duration on students’ academic performance. Liao and Wu (2022) used
multimodal learning analytics combined with quantitative data to analyze the impact of digital
distractions and peer learning on students’ academic performance. In addition, Alshraideh
et al. (2024) and Barata et al. (2016) explored the impact of online gaming on students’ GPA.
They proposed using students’ performance and gaming preferences to predict whether
reducing the amount of time spent playing games serves as a mitigating factor in avoiding a
decline in GPA.
Machine learning algorithms have also been applied to the prediction and analysis of
student learning engagement in higher education institutions. For example, Moubayed et al.
(2021) used unsupervised learning algorithms to cluster students’ behavioral data based on
AEDS online activities and interactions and explored how data mining models can be used to identify
students who perform poorly in terms of learning engagement. Sugden et al. (2021)
investigated the multidimensionality theory of learning engagement based on learning
engagement to assess students’ engagement in online psychology learning activities and used
the k-medoids algorithm to cluster students. Wang (2022) explored deep learning-based
emotion recognition to identify more clearly and accurately the emotional characteristics and
affective attitudes of learners’ engagement in online learning. Ayouni et al. (2021) constructed
an intelligent prediction system to predict students’ learning behavior data based on their
learning behavior data in an online environment. s level of learning engagement. To improve
students’ learning engagement, Wan et al. (2019) explored extracting features from learning
activities and study habits to improve students’ learning engagement by designing
instructional intervention experiments. In addition, machine students play a greater role in
academic performance and learning engagement analysis (So et al., 2023), learning
engagement pattern recognition (So et al., 2023), and student engagement metrics to predict
problematic students (Veerasamy et al., 2022). In summary, machine learning plays a
significant role in higher education. It can predict student grades, analyze the level of learning
engagement, identify influencing factors, and design educational interventions to enhance
learning performance. It helps students recognize and strengthen their self-regulated learning
abilities and assists teachers in identifying students with low levels of learning engagement for
timely intervention. Additionally, it optimizes the allocation of educational resources,
improves overall student learning performance, and provides opportunities for high-quality
development and transformation of schools.

2.2 Applications of machine algorithms in higher education

Multiple machine learning algorithms have been applied to higher education to facilitate the
prediction of academic performance, the identification of student types, or the assessment of at-
risk students. For example, Silva et al. (2024) investigated the use of five classification
algorithms: RF, multilayer perceptron (MLP), SVM, NB, and DT. In addition, an integrated
learning approach was applied to combine these five algorithms to improve the accuracy of
students’ performance prediction. Barata et al. (2016) used a classification algorithm based on
students’ game preference data, including NB, logistic regression (LGR), SVM, DT, SVM, KNN,
and J48 classifiers, and a C4.5 decision tree to identify students’ category types. De Backer et al.
(2022a, b) utilized hierarchical cluster analysis and k-means cluster analysis to identify types of
regulation, analyze students’ metacognitive regulatory behaviors in asynchronous computer-
supported collaborative learning (CSCL), identify regulatory types, and investigate the
relationships between these types and students’ conceptual understanding, motivation, and
self-efficacy. Huang et al. (2023) used machine learning algorithms such as SVM, DT, LGR, and
KNN to construct predictive models for at-risk students. Alshraideh et al. (2024) used a
convolutional neural network (CNN) as the main machine learning algorithm for prediction and
identification tasks. Tseng et al. (2023) used the RF algorithm to construct a predictive model with
the Boruta package for feature selection and the tuneRF package for parameter tuning.
In addition, scholars have also studied the effectiveness of different machine learning
algorithms (rule induction (RI), ANN, KNN, NB, and SVM) by applying and comparing them
in terms of final grades (Asif et al., 2017; Wakelam et al., 2020), students’ characteristics
(Polyzou and Karypis, 2019), classroom attendance (Gray and Perkins, 2019; Wakelam et al.,
2020), course grade data (Yagci, 2022), demographic information (Abu Zohair, 2019; Alturki
et al., 2022), learning activity logs (Murata et al., 2023; Riestra-Gonz�alez et al., 2021), at-risk
students (Chen and Cui, 2020; Huang et al., 2023; Zhang et al., 2022), and other applications.
In summary, various machine learning algorithms have been employed in higher education to
achieve functions such as academic performance prediction, student type identification, and
the mining of students with issues to enhance the teaching standards and management quality
of higher education institutions.
2.3 Purpose and research questions of this systematic review Asian Education
To further understand the application of machine learning in higher education, this study and Development
examines empirical research on the application of machine learning in higher education from Studies
the perspective of machine learning methodology and investigates the status, functions,
algorithms used, and research findings of machine learning applications in higher education.
Specifically, this review focuses on the following five research questions.
RQ1. How have publications on the use of machine learning in higher education evolved?
In which journals have these publications been published? And where are these
publications coming from according to the first author?

RQ2. What are the functions of machine learning applications in higher education?

RQ3. What machine learning algorithms are used to predict student performance, learning
engagement, and self-efficacy analysis in higher education?

RQ4. What is the model evaluation methodology of machine learning applications for
predicting student performance in higher education?

RQ5. What are the results of machine learning applications for predicting student
performance in higher education?

3. Method
This systematic review was conducted employing the PRISMA (Preferred Reporting Items for
Systematic Reviews and Meta-analyses) framework, which is widely recognized for its
rigorous and transparent approach to reviewing the literature (Page et al., 2021). The PRISMA
framework was chosen over other systematic review methodologies due to its comprehensive
guidelines that enhance the replicability and transparency of the review process. It provides a
structured approach to identifying, selecting, and evaluating studies, which ensures a
systematic and transparent review process. The PRISMA framework also offers a clear
framework for reporting systematic reviews, which aligns with our aim to provide a thorough
and transparent analysis of the current state of machine learning applications in higher
education.
Although the PRISMA framework is popular and has been recognized by some researchers
and used in their literature review articles, e.g. Liu (2024), Ouyang et al. (2022), Sarmet et al.
(2023), and Zawacki-Richter et al. (2019), PRISMA has limitations. One major limitation is
that it may not fully handle the complexities and nuances of some research topics, especially in
the field of machine learning where the terminology and algorithms are highly specialized. To
deal with this, we made our search terms as inclusive as possible to cover a wide range of
relevant literature. Also, PRISMA doesn’t specifically guide the selection of search terms or
databases, which might cause differences in search comprehensiveness.
This systematic review aims to discuss the above five hypothetical questions using a
systematic and replicable search strategy. The search strategy identifies inclusion and
exclusion criteria (Ouyang et al., 2022). Based on these criteria, 67 empirical research articles
on the application of machine learning in higher education to predict student performance,
learning engagement, and self-efficacy are mapped in the contribution. The following section
will describe the systematic review procedures.

3.1 Database search

Despite these limitations, the PRISMA framework’s structured approach to identifying,
selecting, and evaluating studies ensured a systematic and transparent review process. We
AEDS believe that the PRISMA’s strengths in providing a clear framework for reporting systematic
reviews outweigh its limitations in our context. The additional measures we have taken help
ensure a comprehensive and rigorous literature review. So, to mitigate this, we selected a
diverse set of reputable databases and used a structured approach to develop our search terms.
To obtain relevant articles, structured search terms were performed on the four international
electronic databases, and specific keywords were used according to the requirements of each
database. The four electronic databases used are Web of Science, ProQuest Central, PubMed,
and IEEE Xplore Digital Library (covering topic, abstract, title/abstract, and full text and
metadata) (see Table 1). In terms of the objectives of this review, the following search terms
were used to obtain data from the four databases. (see Table 2). The search strategy was
conducted on 7 June 2024, with an initial 5,601 records identified.

3.2 Inclusion and exclusion criteria

To achieve the research objective, two criteria were designed to locate relevant articles that
investigated the application of machine learning algorithms in predicting student performance
in higher education. According to the research objective, a set of inclusion and exclusion
criteria was given (see Table 3).
Studies officially published in journals and written in English were included, and articles
such as reviews, conferences, books, news, magazines, data papers, letters, meetings, editorial
materials, and early access or proceeding papers were excluded. In addition, completed studies
whose full texts were available were retrieved, and the participants were students. That is
studies in which participants were teachers or teachers and students were excluded.
Furthermore, studies published are focused only on empirical studies and must focus on the

Table 1. Database sources

Access
No Databases Scope of search date Web site Records

1 Web of Science Topic 7 June, https:// 2,898

2024 www.webofscience.com
2 ProQuest Central Abstract 7 June, http://www.proquest.com 2,154
2024
3 PubMed Title/abstract 7 June, https:// 46
2024 pubmed.ncbi.nlm.nih.gov/
4 IEEE Xplore Digital Full text and 7 June, https://ieeexplore.ieee.org/ 8
Library metadata 2024
Source(s): Authors’ own creation

Table 2. Search terms

Topic Search terms

higher education higher education or university or college or vocational education or higher education
institute or undergrad* or graduate
And machine learning education data mining OR EDM or machine learning OR data mining OR
classification algorithm OR clustering analysis OR association rule learning OR deep
learning OR neural network OR decision trees OR random forest OR support vector
machine OR k-nearest neighbors OR ensemble method
And student student performance prediction, student engagement, or self-efficacy
performance
Source(s): Authors’ own creation
Table 3. Inclusion and exclusion criteria Asian Education
and Development
Inclusion criteria Exclusion criteria Studies
Journal Not Journal (e.g. reviews, conferences, books, news,
magazines, data paper, letters, meetings, editorial
material)
Officially published Unpublished articles (e.g. early access, proceeding
paper)
Written in English Not written in English
The research area is in higher education Not about higher education (e.g. Chemistry,
engineering et al.)
Full text must be retrieved Reports not retrieved (Inaccessible and unavailable
full-text articles)
Participants are the only student Participants are teacher or teacher and student
Scopes are higher education, college, or vacation Not higher education or college or vacation (e.g. k-12,
high school, middle school, adult education, primary
school)
Studies must be conducted about the application of Studies do not conduct about the application of
machine learning to student performance, machine learning to student performance,
engagement, and self-efficacy engagement, and self-efficacy
Empirical studies Studies that do research in abstract studies, theoretical
studies, and review studies
Source(s): Authors’ own creation

application of machine learning to student performance, engagement, and self-efficacy.

Therefore, studies that research theoretical studies and review studies in abstracts are removed.
The criteria of this systematic review focus only on the research progress of higher education,
university, college, vocational education, higher education institutes, or undergraduate or
graduate students; therefore, studies referring to k-12, high school, middle school, adult
education, and primary school are excluded.

3.3 Screening process

The article screening process was carried out via the following steps from the PRISMA flow
chart, as shown in Figure 1. The search retained 5,106 articles from the electronic databases
mentioned earlier by the previously used search term. The flow chart has the following 4 steps:
(1) Remove 139 articles marked as ineligible by the automation tool Zotero. Among them, 6
articles were marked as retracted, and 133 articles were marked as duplicated. (2) 711 articles
whose document type was not an article were removed. (e.g. reviews, conferences, books,
news, magazines, data papers, letters, meetings, editorial materials), 228 unpublished articles
(e.g. early access, proceedings papers) and 46 articles not written in English. After that, we
removed 1,037 articles based on criteria that were not relevant to education (e.g. chemistry,
engineering, etc.). (3) 171 articles that were not retrieved and downloaded due to invalid DOI
links were removed. (4) By reviewing the titles, abstracts, and full texts, a certain number of
articles were excluded based on the eligibility criteria. These included articles were discussed
by the first author and second author to determine whether they were adopted for the aim of the
research. In this step, through reviewing the full text and earlier criteria, the following issues
are discussed: (1) Do the participants in the article relate only to the student? (2) Does the
research scope of the article relate to higher education other than K-12, high school, middle
school, adult education, and primary school? (3) Does the topic addressed in the article relate to
the application of machine learning for predicting student performance, learning engagement,
and self-efficacy in higher education institutions? (4) Is the context of this article an empirical
study? (5) Does the research topic of the articles focus on performance, engagement, or self-
efficacy? In the context of the above issues, we discussed and compared them thoroughly and
AEDS Identification of studies via databases and registers

5106 Records identified from

Identification Web of Science (n = 2898) Records removed before screening:
ProQuest Central (n = 2154) Records marked as ineligible by automation
PubMed (n = 46) tools Zotero: (n = 140)
IEEE (n = 8)

Records excluded with exclusion criteria (n = 2022)

Records screened 1. Not journal: 711
(n = 4967) 2. Unpublished articles: 228
3. Not in English: 46
4. Not education in categories (n =1037)
Screening

Reports sought for retrieval

(n = 2945) Reports not retrieved (n = 171)

Reports excluded: (n = 2706)

1. No about student (n = 63)
Reports assessed for eligibility by 2. No high education (n = 18)
title, abstract and full-text (n = 2774) 3. Not about performance, engagement or self-
efficacy (n = 1063)
4. Not machine learning and prediction (n =
1532)
Included

5. Not empirical study (n = 30)

Reports of included empirical studies
(n = 67)

Source(s): Authors’ own creation, Page et al. (2021)

Figure 1. PRISMA flow chart of the study selection process (The PRISMA, 2020 statement: an updated
guideline for reporting systematic reviews)

finally determined that 67 articles formed the final literature from January 2016 to June 2024
for review analysis.

4. Results and discussions

This literature review systematically analyzes the application of machine learning in the
prediction of learning in the field of higher education research, focusing on the relationships
among student performance prediction, learning engagement, and self-efficacy analysis. This
section presents the results of the research and discusses the findings about the research
objectives of this review.

4.1 RQ1 how have publications on the use of machine learning in higher education
evolved? In which journals have these publications been published? And where are these
publications coming from according to the first author?
4.1.1 Distribution of journals. A total of 67 journal papers were selected for this systematic
literature review. These 67 articles were published in 44 journals, and 50 (approximately
72.06%) of the studies were published in journals with JCRs in Q1 and Q2. This indicates that
the selected literature is representative of the research and that they all delve into the Asian Education
application of machine learning in higher education. Table 4 lists the journals that have and Development
published at least two or more studies on machine learning in higher education concerning Studies
prediction, and in this selected dataset, the top six publications are “Computers and Education”
and “Education and Information Technology”.
4.1.2 Distribution by year. Since 2016, the number of empirical studies related to the
application of machine learning algorithms to predict academic performance has been
increasing. However, before 2018, only a few empirical research articles were published.
Since then, the number of research articles published has increased significantly. In 2022,
when the global COVID-19 epidemic was in a rapid and stable period, the number of research
articles published reached a peak of 21. Figure 2 shows the number of empirical research
articles published annually from 2016 to June 2024 on the prediction of student performance
by machine learning in higher education. The growth direction of the folded line in Figure 2
indicates that machine learning algorithms have high attention and research potential in
predicting student performance, learning engagement, and self-efficacy in future research.
4.1.3 Counties distribution. To clarify the geographical distribution of the research
literature, this study maps the geographical regions of the literature based on the country
information reported by the first authors of the 67 articles through the MapChart website
online. As shown in Figure 3, 28 countries have focused on the application of machine learning

Table 4. Distribution of articles by journal

No Journal name Paper count JCR rank

1 Computers and Education 6 Q1

2 Education and Information Technologies 6 Q1
3 IEEE Transactions on Learning Technologies 5 Q2
4 International Journal of Emerging Technologies in Learning 4 Q2
5 Journal of Educational Computing Research 3 Q1
6 Informatics in Education 2 Q1
7 International Journal of Distance Education Technologies 2 Q1
8 International Journal of Educational Technology in Higher Education 2 Q1
9 Smart Learning Environments 2 Q1
10 Technology Pedagogy and Education 2 Q1
11 Other Journals 33
Note(s): Results from https://jcr.clarivate.com/Search time: 21–06–2024
Source(s): Authors’ own creation

Figure 2. Histogram of a year of publication and number of articles in the literature

AEDS

Figure 3. Geographical distribution of the research literature

in higher education. China, India, Turkey, and Pakistan are leading empirical research on the
application of machine learning in higher education, with a focus on the effects of machine
learning on students’ academic performance, learning engagement, and self-efficacy. China
has the largest number of published articles, with 12 articles (17.91%), followed by India (7,
10.45%), Turkey (6, 8.96%) and Pakistan (5, 7.46%). Additionally, Australia, the United Arab
Emirates, Saudi Arabia, and the United Kingdom followed with three empirical research
articles each. The United States, home to many renowned researchers, has only two articles,
which is the same as the volume of research literature from Germany, Canada, Thailand, and
others. The remaining countries have only one research article each. It is evident that there is a
significant difference in the degree of attention that various countries around the world pay to
the application of machine learning in higher education, and the distribution of research is
highly imbalanced. Nevertheless, researchers from 28 countries are focusing on this field of
study. For example, researchers from Fiji are also paying attention to the application of
machine learning in higher education. This finding indicates that empirical research related to
academic performance in higher education, such as studies on learning engagement and self-
efficacy, is taking shape and may attract the common attention and significant research interest
of researchers, thereby leading to more empirical research outcomes.
Our results highlight a significant imbalance in the geographical distribution of studies
focusing on the application of machine learning in higher education. This section discusses the
potential implications of this imbalance and how future research could address this issue.
Additionally, we analyze whether cultural differences influence the effectiveness of ML
algorithms and the factors predicting student outcomes.
Implications of Geographical Imbalance: The concentration of research in certain regions,
particularly China, India, Turkey, and Pakistan, suggests a potential bias in the generalizability
of findings. This imbalance could be due to varying levels of investment in educational
technology, research capacity, and the prioritization of educational research in different
countries. It may also reflect the digital divide, where countries with more advanced
technological infrastructure are better equipped to conduct and publish research on machine
learning applications in education.
Addressing Geographical Imbalance in Future Research: To address this imbalance, future
research should intentionally target regions that are currently underrepresented in the
literature. International collaborations could be fostered to support research initiatives in Asian Education
countries with emerging educational technology sectors. Additionally, funding agencies and and Development
educational institutions could prioritize research that explores the application of ML in diverse Studies
geographical and cultural contexts.
Cultural Influence on ML Algorithm Effectiveness: Cultural differences may influence the
effectiveness of ML algorithms and the factors predicting student outcomes. Educational
practices, student expectations, and the value placed on different learning outcomes can vary
significantly across cultures. These factors can affect student engagement, motivation, and
performance, which are crucial for the accurate prediction models. For instance, algorithms
trained on data from one cultural context may not perform as well when applied to a different
cultural context due to differing educational norms and student behaviors.
4.1.4 Keyword cloud diagram with keywords. This review generates a keyword cloud
distribution map based on the keyword data provided by the authors in 62 articles, among
which 5 have no keywords. As shown in Figure 4, the most frequently occurring keywords
related to machine learning applications in higher education between 2016 and 2024 appeared
in the articles published during that period. Generally, among the selected keywords, a larger
font size implies a higher frequency of attention and research. Keywords such as “machine”,
“learning”, “data”, “mining”, “student”, “performance”, “prediction”, “classification”,
“decision”, “analytics”, “engagement”, “academic”, and “education” have more prominent
font sizes, being larger than other words, and are hot topics and research foci during this period.
This indicates that in recent studies, researchers have highly emphasized and applied machine
learning algorithms to mine, analyze, and predict the information within educational data, with
a particular focus on data analysis. In terms of machine learning algorithms, the highly noticed
keywords include random, forest, ensemble, clustering, artificial, neural, network, model, etc.
This shows that RF algorithms and ANNs are more commonly used in prediction,
classification, modeling, and evaluation.

4.2 RQ2 what are the functions of machine learning applications in higher education?
Among the articles selected for this study, machine learning in higher education involves
student performance prediction (n 5 55) and student performance analysis (with interventions
for academic performance, learning engagement, and self-efficacy analysis, n 5 12).

Figure 4. Word cloud of machine learning algorithm keywords

AEDS 4.2.1 Prediction of student performance. The first major function of machine learning
applied in higher education is student performance prediction. Among the 55 articles, 3
focused on improving academic performance, 5 focused on analyzing and predicting academic
engagement, and 47 focused on data mining and prediction based on big data in education.
Student performance prediction: In the first classification Nti et al. (2022) collected data
from a total of 550 students from the Sunyani University of Technology in Ghana through
quantitative research methods and questionnaires. Research has revealed that there is a
significant negative relationship between students’ use of social media in the classroom
(multitasking) and their GPA, where the time and frequency of social media use and the nature
of social media are also important factors affecting students’ academic performance. Barata
et al. (2016) collected data from discussion posts on course forums, student engagement in
games, acquisition of achievement badges, and student game preferences through
questionnaire surveys. By combining student performance data and game preferences,
student types can be predicted with up to 79% accuracy in the middle of the semester.
Ramaswami et al. (2019) used student engagement data and self-regulation activity data to
improve the accuracy of student performance prediction. The research results showed that
combining educational data mining techniques with student activity process characteristics
can effectively predict student academic performance. Evangelista and Sy (2022) constructed
a dataset with 649 student records to explore how to improve the predictive performance of
various single classification algorithms by using homogeneous ensemble methods (such as
Bagging and Boosting) and heterogeneous ensemble methods (such as voting and stacking).
The study revealed that the Bagging ensemble method using the OneR algorithm as the base
classifier achieved the best performance in terms of predictive accuracy, with the accuracy rate
increasing from 93.10% to 93.62%.
Data mining and prediction: In the second classification, Moubayed et al. (2021) collected
event log data of students from the learning management system and reported that the number
of logins and the average duration of homework submissions are representative indicators of
student engagement, whereas the number of content readings was not. Wan et al. (2019)
employed methods of data mining, machine learning, and transfer learning to predict student
performance in computer architecture courses via feature engineering. The study revealed that
reminder interventions can significantly improve learner engagement, and error-related
recommendation interventions also play a positive role in enhancing learning engagement. Raj
and VG (2022) collected data from 7,775 students in social science research courses from the
OULAD dataset and used machine learning algorithms for prediction. The results showed that
the RF algorithm performed best in predicting student engagement, with an accuracy of 95%,
precision of 95%, and relevance of 98%. Veerasamy et al. (2022) suggested that results from
continuous assessment tasks could serve as indicators of student engagement and could be
used to identify students in need of support. So et al. (2023) suggested that participation in
generic competence development (GCDA) not only does not negatively affect students’
academic performance but also might help to enhance academic performance, and they
verified the use of genetic algorithms (GAs) and stacking to improve the accuracy of
predicting student engagement. The third classification involves the prediction of student
grades, which is mainly based on educational data such as student learning behavior,
interactive data, academic performance and characteristics, and course grades. Different
machine learning methods are used to predict student performance and different researchers
use different data sources (See Table 5).
Their research revealed that machine learning algorithms play a significant role in
predicting student grades and academic performance for different datasets.
4.2.2 Analyzing student performance. The second major function of machine learning in
higher education is the analysis of student performance. Among these 11 studies, 6 analyzed
the factors affecting the improvement of academic performance and intervention measures,
and 5 analyzed student learning engagement.
Table 5. Methods and results for predicting student performance based on different data sources Asian Education
and Development
Data sources Methods Results References Studies
Learning Researchers commonly The Random Forest Sathe and Adamuthe (2021),
behavior data employed machine learning algorithm demonstrates high Kukkar et al. (2023), Mi et al.
algorithms, particularly accuracy in predicting (2022), Rodrigues et al. (2019),
Random Forest and Support students’ academic Tsiakmaki et al. (2021)
Vector Machine (SVM) as the performance, and feature
core methods in their studies, selection is crucial for
and combined them with other enhancing model
algorithms such as Decision performance, highlighting the
Tree and Naive Bayes, to potential application of
predict students’ academic machine learning techniques
performance in educational data mining
Course grades Researchers commonly Their aggregated findings Alamgir et al. (2024), Alharbi
utilized machine learning and consistently demonstrated et al. (2021), Ali and Hanna
data mining techniques, that course grades are a (2022), Alturki et al. (2022),
focusing on course grades as a significant predictor of Asif et al. (2017), Bertolini et al.
key predictor, to forecast student success. Integrating (2021), Chen and Cui (2020),
student academic performance these grades into machine Goundar et al. (2022), Hussain
learning models substantially et al. (2019), Jamjoom et al.
enhances the accuracy of (2021), Altun et al. (2022), Liu
academic performance (2024), Meghji et al. (2023),
prediction, underscoring the Murata et al. (2023), Nachouki
utility of course grade data in and Abou Naaj (2022),
educational analytics Nuankaew et al. (2020),
Polyzou and Karypis (2019),
Son et al. (2022), Wakelam et al.
(2020), Zhang et al. (2022)
Demographic Researchers commonly Machine learning models, Aggarwal et al. (2021), Alturki
employed machine learning especially those using et al. (2022), Francis and Babu
techniques, particularly ensemble methods and a (2019), Gray and Perkins
classification algorithms like combination of academic and (2019), Hussain et al. (2019),
Random Forest, Support non-academic parameters, Nuankaew et al. (2020), Pardo
Vector Machines, and can significantly predict et al. (2017), Yagci (2022)
Decision Trees, to predict student performance,
student academic performance underscoring the efficacy of
data-driven approaches in
educational analytics for
early identification of at-risk
students
Learning logs Researchers in the file Machine learning models, Hang et al. (2022),
employed similar especially those Trakunphutthirak and Lee
methodologies that included incorporating Random Forest (2022)
the use of machine learning and GBDT, were effective in
algorithms, particularly predicting student
Random Forest and Gradient performance with high
Boosting Decision Tree accuracy, underscoring the
(GBDT), to analyze potential of these algorithms
educational data and predict in educational data mining for
student performance early identification of at-risk
students and enhancement of
academic outcomes

(continued )
AEDS Table 5. Continued

Data sources Methods Results References

Family Researchers commonly Machine learning models, Aggarwal et al. (2021), Zheng
background employed machine learning particularly those and Li (2024)
algorithms, particularly incorporating academic and
ensemble and classification non-academic parameters,
algorithms such as Random significantly enhance the
Forest, Support Vector accuracy of student
Machines, and Naive Bayes, to performance prediction, with
build and evaluate models random forest algorithms
predicting students’ academic consistently demonstrating
performance high predictive accuracy
across studies
Interaction Researchers commonly Their collective findings Aydog �du (2020), Bere et al.
data utilized machine learning show that Machine learning (2022), Liu et al. (2022),
techniques, particularly models, especially those Riestra-Gonz�alez et al. (2021),
classification algorithms like incorporating ensemble Xue and Niu (2023), Y€ ur€um
Random Forest, Support methods like Random Forest, et al. (2023)
Vector Machines, and Naive can significantly predict
Bayes, to predict student student performance with
academic performance high accuracy, underscoring
the value of these algorithms
in educational data mining for
early identification of at-risk
students
Others Researchers employed similar Their collective findings Durak and Bulut (2024),
methodologies, consistently demonstrated the Parhizkar et al. (2023), Mushi
predominantly utilizing efficacy of machine learning and Ngondya (2021), Sultana
machine learning algorithms in predicting academic et al. (2017)
such as Decision Trees, performance, with a
Random Forests, and Support particular emphasis on the
Vector Machines to predict significant role of algorithm
student performance selection and feature
engineering in enhancing
prediction accuracy
Source(s): Authors’ own creation

Student performance: In the first classification, Huang et al. (2023) concluded that
personalized interventions not only improved students’ academic performance but also
promoted their competence in the following learning strategies: rehearsal, critical thinking,
metacognitive self-regulation, effort regulation, and peer learning. The results of Tseng et al.
(2023) showed that students with personalized learning performed significantly better than
other students did. Silva et al. (2024) reported that the MLP algorithm had the highest accuracy
and the lowest error value in the classification of factors affecting students’ performance. Liao
and Wu (2022) analyzed learner discourse data and reported that the predictive validity of peer
learning engagement for academic performance was better than that of self-perceived peer
learning orientation. Alshraideh et al. (2024) analyzed historical data on students’ online game
participation and reported that 80% of students who consistently played online games daily
experienced a decrease in GPA. De Backer et al. (2022a, b) collected online discussion
records, learning motivations, and self-efficacy data (obtained through questionnaires) from
196 first-year university students to explore how students’ metacognitive regulation behaviors
affect their academic performance in a computer-supported collaborative learning (CSCL)
environment. The results revealed significant differences in conceptual understanding,
learning motivation, and self-efficacy among comprehensive affirmational regulators
(AOAR), social-oriented elaborative regulators (SOER), and individual-oriented passive
regulators (IOPR).
 Student learning engagement: In the second category, Wu (2021) collected information on Asian Education
78 students learning an advanced statistics course, including demographic data, motivational and Development
characteristics, and course grades. The study results indicated that help-seeking behavior Studies
positively predicted course grades, whereas ICT-related academic procrastination negatively
predicted course grades. Wang (2022) collected text information on learning comments and
interactive behaviors from an online learning platform and reported that deep learning-based
methods could effectively identify the emotions associated with college students’ engagement
in online learning. Ayouni et al. (2021) analyzed reports on student logins, participation in
forums and group activities, and access to course materials and reported that measuring student
engagement and providing timely feedback could significantly improve students’ learning
engagement. The study also revealed that the artificial neural network (ANN) performed best
in predicting student engagement, with an accuracy rate of 85%.

4.3 RQ3 what machine learning algorithms are used to predict student performance,
learning engagement, and self-efficacy analysis in higher education?
Among the 67 articles, each article clearly identified the machine learning algorithms they
used in their research, and most of the articles used multiple machine learning algorithms.
According to the results of the literature review, 67 articles shared 41 algorithms, with a total
frequency of algorithm usage of 256. Among them, the most used machine learning algorithms
are the DT (n 5 39, 15.29%), RF (n 5 36, 14.12%), NB (n 5 25, 98.0%), and SVM (n 5 25,
98.0%) algorithms. The specific frequency of algorithm usage is shown in Figure 5.
In addition, some studies have used up to 9 types of algorithms in one study (see Table A1 in
the Appendix).
Decision Trees (DT), Random Forests (RF), Naı€ve Bayes (NB), and Support Vector
Machines (SVM) were the most frequently utilized ML algorithms. These algorithms were
chosen for their ability to handle various data types and structures, from the binary splits of DT
to the ensemble learning capabilities of RF, which generally outperformed single algorithms in
terms of accuracy and other evaluation metrics. Additionally, K-means and advanced machine
learning algorithms, such as NN, KNN, ANN, DNN, and RNN, have also been researched
multiple times. For example, Asif et al. (2017), Gray and Perkins (2019) and Wakelam et al.
(2020) used data mining techniques such as DT, RF, and NB to analyze and predict
undergraduate students’ performance. Polyzou and Karypis (2019) predicted
underperforming students in the next semester through student characteristics via DT, RF,
SVM, and GB. Trakunphutthirak and Lee (2022) used not only DT, RF, and LR algorithms but
also advanced machine learning algorithms such as KNN, GB, and NN. In addition, scholars
have used lesser-known machine learning algorithms, such as LADTree (Son et al., 2022),
OneR (Evangelista and Sy, 2022), Jrip (Evangelista and Sy, 2022; Meghji et al., 2023), ENR
(Bertolini et al., 2021), BR (Abdullah et al., 2023), SVD, NMF and BR (Alharbi et al,. 2021),
FCM (Tsiakmaki et al., 2021), and Kstar (Meghji et al., 2023). Robust (Y€ um et al., 2023),
ur€
Voting and Bagging (Aggarwal et al., 2021) LightGBM (Xue and Niu, 2023), Bayes Net
(Alturki et al., 2022), k-medoids (Sugden et al., 2021), stacking (So et al., 2023), EM (Ali and
Hanna, 2022), the improved linear discriminant analysis technique (Alharbi et al., 2021) and
CA (So et al., 2023).

4.4 RQ4 what is the model evaluation methodology of machine learning applications for
predicting student performance in higher education?
Common model evaluation metrics included accuracy, F1 score, recall, and precision. These
metrics are crucial for assessing the performance of ML models, with accuracy measuring the
proportion of correct predictions, the F1 score balancing precision and recall, and recall and
precision focusing on the model’s ability to find all relevant instances and the accuracy of
positive predictions, respectively. The metrics for evaluating different models constructed via
different ML algorithms have been reported in the literature. A review of the model evaluation
AEDS

Figure 5. Distribution of machine learning algorithms

methods and metrics used in the literature reveals that most research articles have used Asian Education
multiple model evaluation metrics. According to the results of the literature review, a total of and Development
31 different evaluation metrics were used in 67 articles, with a total frequency of 229. The top 5 Studies
most frequently used model evaluation indicators were the accuracy (n 5 38, 16.59%), F1
score (n 5 31, 13.54%), recall (n 5 30, 13.10%), precision (n 5 28, 12.23%), cross-validation
(n 5 16, 6.99%), and AUC (n 5 16, 6.99%). In addition, a study used a maximum of 8 model
evaluation metrics (see Figure 6 and Table A2 in the Appendix).
The evaluation indicators with a relatively high frequency of use have attracted the
attention of most researchers, and multiple evaluation indicators have also been used in a
single literature study. For example, Huang et al. (2023) used AC, precision, recall, F1, AUC,
CV, and CR to evaluate models constructed via machine learning algorithms, and other
researchers used the following indicators: Hussain et al. (2019, 2022), Kukkar et al. (2023),
Alturki and Alturki (2021) and Trakunphutthirak and Lee (2022). In addition to the commonly
used evaluation indicators mentioned above, some researchers have innovatively used
indicators with a lower frequency of use, such as 95% CI (Huang et al., 2023), PTA (Alharbi
et al., 2021), SC and T (Moubayed et al., 2021), MWU (Tseng et al., 2023), α (Wu, 2021),
MANOVA (Moubayed et al., 2021), JC (Bertolini et al., 2021), and SI (Abdullah et al., 2023),
and some researchers have even used Pearson and Spearman (Abdullah et al., 2023) to
evaluate model indicators.

4.5 RQ5 what are the results of machine learning applications for predicting student
performance in higher education?
The findings from the 67 empirical studies collected indicate that machine learning algorithms
exhibit high-quality performance in predicting academic performance, analyzing learning
engagement, and self-efficacy. In this section, we expand on the discussion of these findings,
particularly focusing on the comparative effectiveness of different ML algorithms.
Predicting academic performance: For example, individual algorithms such as RF, SVM,
NB, J48, and MLP perform well in terms of precision, recall, and F1 score during model
evaluation. However, when ensemble learning methods are used in combination with the
aforementioned methods, the predictive results demonstrate that ensemble learning methods
have higher accuracy and lower error rates in predicting students’ academic achievements,
confirming the effectiveness of combining multiple algorithms for prediction, as shown by
Silva et al. (2024). This research finding is highly consistent with the results of Sathe and
Adamuthe (2021). The results of Nti et al. (2022) further indicate that in terms of model
performance, the RF outperforms the DT in predicting students’ academic performance, which
corroborates the findings of Alamgir et al. (2024) and Wu (2021). However, the results of Bere
et al. (2022) show that the KNN algorithm has greater operational efficiency and predictive
effectiveness (such as accuracy, precision, and F1 score) in handling student performance
prediction than the RF and MLP algorithms do.
Learning engagement analysis: Personalized intervention methods not only improved
students’ learning performance but also promoted the enhancement of their abilities in various
learning strategies, such as rehearsal, critical thinking, metacognitive self-regulation, effort
regulation, and peer learning (Huang et al., 2023). Tseng et al. (2023) reported that
personalized learning materials also have a significant effect on improving students’ academic
performance. Sugden et al. (2021) found through experimental and control group analyses that
students have a positive attitude toward online activities, with high engagement, promoting
deep emotional, cognitive, and behavioral involvement in learning. The experimental results
of Ayouni et al. (2021) showed that the proposed intelligent prediction system can effectively
predict and improve students’ online learning engagement. They applied three machine
learning algorithms (DT, SVM, and ANN) to analyze student activity data and reported that the
ANN performed best in predicting student engagement, with an accuracy rate of 85%,
followed by SVM (80%) and DT (75%). Wan et al. (2019) found through the construction of a
AEDS Times and Frequency of Evaluation Methodology in All Studies
Times Frequency
AC 38 16.59%
F1 31 13.54%
Recall 30 13.10%
Precision 28 12.23%
CV 16 6.99%
AUC 16 6.99%
RMSE 10 4.37%
MAE 8 3.49%
CMT 7 3.06%
Kappa 6 2.62%
MSE 5 2.18%
SP 4 1.75%
MAPE 4 1.75%
Assessment indicators

R2 3 1.31%
ANCOVA 2 0.87%
OOB 2 0.87%
KR20 2 0.87%
R 2 0.87%
HO 2 0.87%
SD 2 0.87%
Spearman 1 0.44%
Pearson 1 0.44%
SI 1 0.44%
JC 1 0.44%
MANOVA 1 0.44%
α 1 0.44%
T 1 0.44%
MWU 1 0.44%
SC 1 0.44%
PTA 1 0.44%
95% CI 1 0.44%

0 5 10 15 20 25 30 35 40 45
Number of times the assessment of indicators was used

Note(s): 95% confidence interval: 95% CI, percentage of tick accuracy: PTA, silhouette
coefficient: SC, Mann ‒ Whitney U: MWU, T-test: T, Cronbach's alpha: α, multivariate
analysis of variance: MANOVA, Jaccard coefficient: JC, scatter index: SI, Pearson
coefficient: Pearson, Spearman coefficient: Spearman, standard deviation: SD, hold out:
HO, correlation coefficient: R, Kuder ‒ Richardson20: KR20, out ‒ of ‒ bag: OOB, analysis
of covariance: ANCOVA, R-squared: R2, mean of average precision: MAPE, specificity:
SP, mean square error: MSE, kappa coefficient: kappa, confusion matrix table: CMT, mean
absolute error: MAE, root mean square error: RMSE, area under the curve: AUC,
cross-validation: CV, precision: precision, recall: recall, F1 measure: F1, accuracy: AC
Source(s): Authors’ own creation
Figure 6. Distribution of evaluation methodologies

prediction model that behavior reminders based on early prediction and error-related
recommendation interventions both played a positive role in improving mixed learning
engagement. Raj and VG (2022) constructed a student virtual learning environment
engagement prediction model, and the experimental results revealed that the RF
classification algorithm performed best in predicting student engagement, with an accuracy
rate of 95%, precision of 95%, and recall rate of 98%. Nkomo and Nat (2021) used cluster Asian Education
analysis to divide students’ learning participation into three different levels of engagement, and Development
namely, low, medium, and high, and an experiment revealed a positive correlation between Studies
learning engagement and academic performance.
Self-efficacy analysis: Jamjoom et al. (2021) used DT, KNN, NB, and SVM to construct
indicators of self-efficacy for computer science students, and the study results revealed that DT
and SVM performed best in terms of prediction accuracy, with an accuracy rate of 99.18%,
whereas the accuracy rates of KNN and NB were also relatively high, at 98.36% and 96.72%,
respectively. These results confirm the reliability and effectiveness of the selected classifiers in
predicting student performance, and the results emphasize a strong positive correlation
between self-efficacy and student performance in programming courses. Previous studies
have also revealed that predictions based on indicators of student self-efficacy are more
realistic and effective than those based on demographic characteristics. De Backer et al.
(2022a, b) found through cluster analysis that among the three regulators of comprehensive
affirmational regulators (AOAR), social-oriented elaborative regulators (SOER), and
individual-oriented passive regulators (IOPR), SOER scored the highest in self-efficacy
(M 5 4.21; SD 5 0.31), whereas IOPR scored the lowest in self-efficacy (M 5 2.39;
SD 5 0.31), and the experimental results emphasized the importance of providing customized
metacognitive support in the CSCL environment for different types of regulators. For more
details on the findings and results of each study, see Table A3 in the Appendix.

4.6 Potential biases

In conducting this systematic review, we are cognizant of potential biases that may have
influenced our findings. These biases can arise from various aspects of the study selection and
reporting process.
4.6.1 Selection bias. Selection bias may have occurred if certain types of studies were over-
or under-represented in our review. For instance, our focus on English-language journals could
have led to the exclusion of relevant research published in other languages, potentially missing
diverse perspectives and findings from non-English speaking regions. Additionally, our
reliance on specific databases might have resulted in the omission of studies published in less
indexed or less prestigious outlets, which could contain valuable but less accessible
information.
4.6.2 Reporting bias. Reporting bias might have affected our review if the studies we
included were more likely to report positive results or significant findings, while those with
null or negative results were less likely to be published or included in our analysis. This bias
could skew our understanding of the true effectiveness of ML algorithms in higher education.
4.6.3 Mitigating biases in future research. To mitigate these biases in future research, we
suggest the following strategies:
Inclusive Language Policy: Future reviews should aim to include studies published in
multiple languages to capture a more global perspective on the use of ML in higher education.
Collaborations with international researchers or the use of translation services could help
access non-English studies.
Broad Database Search: Expanding the search to include a wider range of databases,
including those that index grey literature, could help reduce selection bias by ensuring a more
comprehensive capture of relevant research.
Transparent Reporting Standards: Encouraging authors to adhere to transparent reporting
standards, such as the CONSORT (Consolidated Standards of Reporting Trials) or PRISMA
guidelines, could help minimize reporting bias by ensuring that all outcomes, regardless of
significance, are reported.
Protocol Registration: Registering review protocols in advance can help reduce bias by
ensuring that the review process is systematic and less susceptible to changes based on the
results of included studies.
AEDS Sensitivity Analyses: Conducting sensitivity analyses to assess the robustness of findings to
different inclusion criteria can provide insights into the potential impact of selection and
reporting biases on the review’s conclusions.
By acknowledging and addressing these potential biases, we can work towards more
accurate and reliable systematic reviews that better inform educational practice and policy.

5. Comparison with other systematic reviews

Our systematic literature review offers a comprehensive analysis of the application of machine
learning in higher education, focusing on student performance prediction, learning
engagement, and self-efficacy. In this section, we compare our findings with those of other
systematic reviews in the field to highlight similarities and differences, and to discuss how our
review extends or challenges existing literature.
Similarities with Existing Literature: Similar to previous reviews, such as the work by Luan
and Tsai (2021) and Andrade-Gir�on et al. (2023), we found that machine learning algorithms,
particularly ensemble methods like Random Forest, demonstrate high accuracy in predicting
student academic performance. This aligns with the consensus in the literature that machine
learning is a powerful tool for educational data mining and analytics.
Differences from Existing Literature: Where our review diverges is in the depth of analysis
regarding the application of machine learning in the context of learning engagement and self-
efficacy (e.g. Nti et al., 2022; Barata et al., 2016 and Evangelista and Sy, 2022). While other
reviews have focused predominantly on academic performance prediction, we have extended
the scope to include a broader range of student outcomes. This approach provides a more
holistic view of student success in higher education, acknowledging that performance is
influenced by, and influences, engagement and self-efficacy.
How Our Review Extends Existing Literature Our review extends the existing literature by
systematically examining a broader range of machine learning algorithms and their application
across various educational outcomes. We have also included a more diverse set of studies,
covering a wider geographical spread and a larger time frame (2016–2024), which allows for a
more comprehensive understanding of the evolution and current state of machine learning
applications in higher education.
Challenges to Existing Literature: One of the challenges our review presents to existing
literature is the identification of the geographical imbalance in research focus. While previous
reviews may have overlooked this aspect, we highlight the concentration of studies in certain
regions and the potential impact this has on the generalizability of findings. This insight
prompts a call for more diverse and inclusive research in the field.
New Insights Provided: The new insights provided by our review include the significant
role of ensemble learning methods in enhancing prediction accuracy, the importance of
considering cultural differences in algorithm effectiveness, and the potential for machine
learning to inform personalized education and early intervention strategies. Our analysis of the
geographical distribution of research and the emphasis on the implications of this imbalance
also offer new directions for future research.
In conclusion, our systematic literature review builds upon and extends existing literature
by providing a more comprehensive and nuanced understanding of the application of machine
learning in higher education. It challenges the field to consider a broader range of outcomes, to
address geographical imbalances, and to explore the potential of machine learning in
enhancing student engagement and self-efficacy alongside academic performance.

6. Conclusion and future searches

This article provides a systematic review of empirical studies on the application of machine
learning in higher education, primarily examining publication journals, geographical
distributions, publication time distributions, functions, algorithms used, model evaluation
indicators, and the results and effectiveness of empirical studies. Although empirical research Asian Education
on machine learning in higher education is limited, it has proven that there are differences and Development
between machine learning algorithms, but they exhibit high accuracy, high precision, and high Studies
recall in model prediction and evaluation, which is beneficial for ensuring the credibility of
student grade prediction, learning engagement, and self-efficacy analysis results (Liao and
Wu, 2022; Nti et al., 2022; Silva et al., 2024). This helps to transform teaching design and
teaching models in higher education institutions, develop teaching methods, promote the
active role of machine learning in personalized student training (Tseng et al., 2023), promote
early intervention in student learning behavior (Huang et al., 2023), and identify and solve real
problems faced by students (Alshraideh et al., 2024), thereby enhancing the teaching level and
quality of higher education.

6.1 Limitations of the study

Although before the literature search, this review carefully screened the search terms, created a
list of terms, and established criteria for the inclusion and exclusion of literature, it is not
possible to ensure the completeness of the literature search and the bias in the exclusion of
literature. Several limitations must be acknowledged. Firstly, the selection of databases was
limited to four well-established electronic databases, which may have excluded relevant
studies published in other forums or databases. This could potentially introduce a bias towards
studies that are more likely to be indexed in these databases. Secondly, by focusing on journal
articles, our review may have missed important contributions from conference proceedings,
grey literature, and other non-peer-reviewed sources that could offer additional insights.
Additionally, the exclusion of non-English studies limits the generalizability of our findings to
non-English speaking regions and may overlook valuable research conducted in other
languages.

6.2 Directions for future research

Investigating the Effectiveness of ML Algorithms in Different Subject Areas: Future research
should investigate the effectiveness of ML algorithms in predicting student outcomes across
different subject areas. This could help identify subject-specific factors that influence
algorithm performance and adapt models to better suit the unique characteristics of different
disciplines.
Exploring Unsupervised and Self-Supervised Learning Methods: While our review
focused on supervised learning methods, there is a need to explore the potential of
unsupervised and self-supervised learning methods in the context of educational data. These
methods could uncover hidden patterns and structures in data without the need for labeled
outcomes, potentially leading to new insights into student behavior and performance.
Developing Interventions Based on ML Predictions: An important direction for future
research is the development and evaluation of educational interventions based on predictions
made by ML models. This would involve not just predicting outcomes but also testing
strategies to improve them, providing a more direct test of the practical utility of ML in
enhancing student learning and success.
Addressing Geographical and Cultural Variability: Future research should aim to address
the geographical and cultural imbalance in the current literature by conducting studies in
diverse regions and cultural contexts. This could involve international collaborations and the
use of culturally adaptive ML models that can account for variations in educational practices
and student behaviors across different settings.
Enhancing Model Generalizability: Research should focus on enhancing the
generalizability of ML models by testing their performance across different datasets and
educational contexts. This could involve cross-validation techniques and the development of
models that are robust enough to handle variations in data quality and structure.
AEDS Incorporating Non-English Studies: To broaden the scope of future reviews, efforts should
be made to include non-English studies, possibly through translation or collaboration with
researchers in those regions. This would provide a more global perspective on the application
of ML in higher education.
In conclusion, while our review provides a thorough analysis of the current state of machine
learning applications in higher education, there are clear opportunities for future research to
build upon these findings. By addressing the limitations identified and pursuing the directions
for future research proposed, we can continue to advance the understanding and effective
application of ML in educational settings.

References
Abdullah, M., Al-Ayyoub, M., Shatnawi, F., Rawashdeh, S. and Abbott, R. (2023), “Predicting
students’ academic performance using e-learning logs”, IAES International Journal of Artificial
Intelligence, Vol. 12 No. 2, p. 831, doi: 10.11591/ijai.v12.i2.pp831-839.
Abu Zohair, L. (2019), “Prediction of Student’s performance by modelling small dataset size”,
International Journal of Educational Technology in Higher Education, Vol. 16 No. 1, 27, doi:
10.1186/s41239-019-0160-3.
Aggarwal, D., Mittal, S. and Bali, V. (2021), “Identifying non-performing students in higher
educational institutions using data mining techniques”, International Journal of Information
System Modeling and Design, Vol. 12 No. 1, pp. 94-110, doi: 10.4018/IJISMD.2021010105.
Alamgir, Z., Akram, H., Karim, S. and Wali, A. (2024), “Enhancing student performance prediction via
educational data mining on academic data”, Informatics in Education, Vol. 23 No. 1, pp. 1-24,
doi: 10.15388/infedu.2024.04.
Alharbi, B., Assiri, F. and Alharbi, B. (2021), “A comparative study of student performance prediction
using pre-course data”, ADCAIJ: Advances in Distributed Computing and Artificial Intelligence
Journal, Vol. 10 No. 1, pp. 49-61, doi: 10.14201/ADCAIJ20211014961.
Ali, A. and Hanna, W. (2022), “Predicting students’ achievement in a hybrid environment through self-
regulated learning, log data, and course engagement: a data mining approach”, Journal of
Educational Computing Research, Vol. 60 No. 4, pp. 960-985, doi: 10.1177/
07356331211056178.
Alshraideh, M., Abu-Zayed, A. A.-J., Leiner, M. and AlDajani, I.M. (2024), “Beyond the scoreboard: a
machine learning investigation of online games’ influence on Jordanian university students’
grades”, Applied Computational Intelligence and Soft Computing, Vol. 2024, pp. 1-11, doi:
10.1155/2024/1337725.
Altun, M., Kayıkçı, K. and Irmak, S. (2022), “A model proposal for predicting students’ academic
performances based on data mining”, Hacettepe University Journal of Education, Vol. 37 No. 3,
pp. 1080-1098, doi: 10.16986/HUJE.2021068491.
Alturki, S. and Alturki, N. (2021), “Using educational data mining to predict students’ academic
performance for applying early interventions”, Journal of Information Technology Education:
Innovations in Practice, Vol. 20, pp. 121-137, doi: 10.28945/4835.
Alturki, S., Cohausz, L. and Stuckenschmidt, H. (2022), “Predicting Master’s students’ academic
performance: an empirical study in Germany”, Smart Learning Environments, Vol. 9 No. 1, p.
38, doi: 10.1186/s40561-022-00220-y.
Andrade-Gir�
on, D., Sandivar-Rosas, J., Mar�ın-Rodriguez, W., Ramirez, E., Toro-Dextre, E., Ausejo-
Sanchez, J., Villarreal-Torres, H. and Angeles-Morales, J. (2023), “Predicting student dropout
based on machine learning and deep learning: a systematic review”, Eai Endorsed Transactions
on Scalable Information Systems, Vol. 10 No. 5, doi: 10.4108/eetsis.3586.
Asif, R., Merceron, A., Ali, S.A. and Haider, N.G. (2017), “Analyzing undergraduate students’
performance using educational data mining”, Computers and Education, Vol. 113, pp. 177-194,
doi: 10.1016/j.compedu.2017.05.007.
Aydog�du, Ş. (2020), “Predicting student final performance using artificial neural networks in online Asian Education
learning environments”, Education and Information Technologies, Vol. 25 No. 3, pp. 1913-1927, and Development
doi: 10.1007/s10639-019-10053-x. Studies
Ayouni, S., Hajjej, F., Maddeh, M. and Al-Otaibi, S. (2021), “A new ML-based approach to enhance
student engagement in online environment”, PLoS One, Vol. 16 No. 11, e0258788, doi: 10.1371/
journal.pone.0258788.
Baker, R.S. and Yacef, K. (2009), “The state of educational data mining in 2009: a review and future
visions”, Journal of Educational Data Mining, Vol. 1 No. 1, pp. 3-17, doi: 10.5281/
zenodo.3554657.
Barata, G., Gama, S., Jorge, J. and Gonçalves, D. (2016), “Early prediction of student profiles based on
performance and gaming preferences”, IEEE Transactions on Learning Technologies, Vol. 9
No. 3, pp. 272-284, doi: 10.1109/TLT.2016.2541664.
Bere, S.S., Shukla, G.P., Khan, V.N., Shah, A.M. and Takale, D.G. (2022), Analysis of students
performance prediction in online courses using machine learning algorithms, Vol. 20 No. 12.
Bertolini, R., Finch, S. and Nehm, R. (2021), “Enhancing data pipelines for forecasting student
performance: integrating feature selection with cross-validation”, International Journal of
Educational Technology in Higher Education, Vol. 18 No. 1, 44, doi: 10.1186/s41239-021-
00279-6.
Chen, F. and Cui, Y. (2020), “Utilizing student time series behaviour in learning management systems
for early prediction of course performance”, Journal of Learning Analytics, Vol. 7 No. 2,
pp. 1-17, doi: 10.18608/jla.2020.72.1.
De Backer, L., Van Keer, H., De Smedt, F., Merchie, E. and Valcke, M. (2022a), “Identifying regulation
profiles during computer-supported collaborative learning and examining their relation with
students’ performance, motivation, and self-efficacy for learning”, Computers and Education,
Vol. 179, 104421, doi: 10.1016/j.compedu.2021.104421.
De Backer, L., Van Keer, H. and Valcke, M. (2022b), “The functions of shared metacognitive
regulation and their differential relation with collaborative learners’ understanding of the
learning content”, Learning and Instruction, Vol. 77, doi: 10.1016/j.learninstruc.2021.101527.
Durak, A. and Bulut, V. (2024), “Classification and prediction-based machine learning algorithms to
predict students’ low and high programming performance”, Computer Applications in
Engineering Education, Vol. 32 No. 1, e22679, doi: 10.1002/cae.22679.
Evangelista, E.D.L. and Sy, B.D. (2022), “An approach for improved students’ performance prediction
using homogeneous and heterogeneous ensemble methods”, International Journal of Electrical
and Computer Engineering, Vol. 12 No. 5, p. 5226, doi: 10.11591/ijece.v12i5.pp5226-5235.
Fahd, K., Venkatraman, S., Miah, S. and Ahmed, K. (2022), “Application of machine learning in higher
education to assess student academic performance, at-risk, and attrition: a meta-analysis of
literature”, Education and Information Technologies, Vol. 27 No. 3, pp. 3743-3775, doi:
10.1007/s10639-021-10741-7.
Francis, B.K. and Babu, S.S. (2019), “Predicting academic performance of students using a hybrid data
mining approach”, Journal of Medical Systems, Vol. 46 No. 6, p. 162, doi: 10.1007/s10916-019-
1295-4.
Goundar, S., Deb, A., Lal, G. and Naseem, M. (2022), “Using online student interactions to predict
performance in a first-year computing science course”, Technology Pedagogy and Education,
Vol. 31 No. 4, pp. 451-469, doi: 10.1080/1475939X.2021.2021977.
Gray, C. and Perkins, D. (2019), “Utilizing early engagement and machine learning to predict student
outcomes”, Computers and Education, Vol. 131, pp. 22-32, doi: 10.1016/
j.compedu.2018.12.006.
Hang, H., Shuang, D., Jiarou, L. and Zhonglin, K. (2022), Towards a predict on model of learning
perfor- mance: informed by learning behavior Big data analytics.
Hooda, M., Rana, C., Dahiya, O., Shet, J. and Singh, B. (2022), “Integrating LA and EDM for
improving students success in higher education using FCN algorithm”, Mathematical Problems
in Engineering, Vol. 2022, pp. 1-12, doi: 10.1155/2022/7690103.
AEDS Huang, A.Y.Q., Chang, J.W., Yang, A.C.M., Ogata, H., Ting, S., Yen, R.X. and Yang, S.J.H. (2023),
Personalized intervention based on the early prediction of at-risk students to improve their
learning performance, Educational Technology and Society, Vol. 26 No. 4, pp. 69-89.
�lu, F. and Hazarika, G.C. (2019),
Hussain, S., Muhsion, Z.F., Salal, Y.K., Theodoru, P., Kurtog
“Prediction model on student performance based on internal assessment using deep learning”,
International Journal of Emerging Technologies in Learning (iJET), Vol. 14 No. 08, p. 4, doi:
10.3991/ijet.v14i08.10001.
Hussain, A., Khan, M. and Ullah, K. (2022), “Student’s performance prediction model and affecting
factors using classification techniques”, Education and Information Technologies, Vol. 27 No. 6,
pp. 8841-8858, doi: 10.1007/s10639-022-10988-8.
� c, N., Had�zi�c, A. and Jianu, A. (2021), “Needs and performance
Ili�c, M.P., P�aun, D., Popovi�c Sevi�
analysis for changes in higher education and implementation of artificial intelligence, machine
learning, and extended reality”, Education Sciences, Vol. 11 No. 10, p. 568, doi: 10.3390/
educsci11100568.
Jamjoom, M., Alabdulkreem, E., Hadjouni, M., Karim, F. and Qarh, M. (2021), “Early prediction for
at-risk students in an introductory programming course based on student self-efficacy”,
Informatica, Vol. 45 No. 6, doi: 10.31449/inf.v45i6.3528.
Kaddoura, S., Popescu, D.E. and Hemanth, J.D. (2022), “A systematic review on machine learning
models for online learning and examination systems”, PeerJ Computer Science, Vol. 8, p. e986,
doi: 10.7717/peerj-cs.986.
Kukkar, A., Mohana, R., Sharma, A. and Nayyar, A. (2023), “Prediction of student academic
performance based on their emotional wellbeing and interaction on various e-learning
platforms”, Education and Information Technologies, Vol. 28 No. 8, pp. 9655-9684, doi:
10.1007/s10639-022-11573-9.
Liao, C. and Wu, J. (2022), “Deploying multimodal learning analysis models to explore the impact of
digital distraction and peer learning on student performance”, Computers and Education,
Vol. 190, 104599, doi: 10.1016/j.compedu.2022.104599.
Liu, W. (2019), “An improved back-propagation neural network for the prediction of college students’
English performance”, International Journal of Emerging Technologies in Learning (iJET),
Vol. 14 No. 16, p. 130, doi: 10.3991/ijet.v14i16.11187.
Liu, T. (2024), Public health education using social learning theory: a systematic scoping review, doi:
10.1186/s12889-024-19333-9.
Liu, Y., Fan, S., Xu, S., Sajjanhar, A., Yeom, S. and Wei, Y. (2022), “Predicting student performance
using clickstream data and machine learning”, Education Sciences, Vol. 13 No. 1, p. 17, doi:
10.3390/educsci13010017.
Luan, H. and Tsai, C. (2021), “A review of using machine learning approaches for precision
education”, Educational Technology and Society, Vol. 24 No. 1, pp. 250-266.
Meghji, A.F., Mahoto, N.A., Asiri, Y., Alshahrani, H., Sulaiman, A. and Shaikh, A. (2023), “Early
detection of student degree-level academic performance using educational data mining”, PeerJ
Computer Science, Vol. 9 No. e1294, e1294, doi: 10.7717/peerj-cs.1294.
Mi, H., Gao, Z., Zhang, Q. and Zheng, Y. (2022), “Research on constructing online learning
performance prediction model combining feature selection and neural network”, International
Journal of Emerging Technologies in Learning (iJET), Vol. 17 No. 07, pp. 94-111, doi: 10.3991/
ijet.v17i07.25587.
Moubayed, A., Injadat, M., Shami, A. and Lutfiyya, H. (2021), “Student engagement level in
e-learning environment: clustering using K-means”, doi: 10.31235/osf.io/ecg4x.
Murata, R., Okubo, F., Minematsu, T., Taniguchi, Y. and Shimada, A. (2023), “Recurrent neural
network-FitNets: improving early prediction of student performanceby time-series knowledge
distillation”, Journal of Educational Computing Research, Vol. 61 No. 3, pp. 639-670, doi:
10.1177/07356331221129765.
Mushi, P.K. and Ngondya, D. (2021), “Prediction of mathematics performance using educational data Asian Education
mining techniques”, International Journal of Advanced Computer Research, Vol. 11 No. 56, pp. and Development
83-102, doi: 10.19101/IJACR.2021.1152024. Studies
Nachouki, M. and Abou Naaj, M. (2022), “Predicting student performance to improve academic
advising using the random forest algorithm”, International Journal of Distance Education
Technologies, Vol. 20 No. 1, pp. 1-17, doi: 10.4018/IJDET.296702.
Nkomo, L.M. and Nat, M. (2021), “Student engagement patterns in a blended learning environment: an
educational data mining approach”, TechTrends, Vol. 65 No. 5, pp. 808-817, doi: 10.1007/
s11528-021-00638-0.
Nti, I.K., Akyeramfo-Sam, S., Bediako-Kyeremeh, B. and Agyemang, S. (2022), “Prediction of social
media effects on students’ academic performance using Machine Learning Algorithms (MLAs)”,
Journal of Computers in Education, Vol. 9 No. 2, pp. 195-223, doi: 10.1007/s40692-021-
00201-z.
Nuankaew, P., Nuankaew, W., Teeraputon, D., Phanniphong, K. and Bussaman, S. (2020), “Prediction
model of student achievement in business computer disciplines”, International Journal of
Emerging Technologies in Learning (iJET), Vol. 15 No. 20, p. 160, doi: 10.3991/
ijet.v15i20.15273.
Okewu, E., Adewole, P., Misra, S., Maskeliunas, R. and Damasevicius, R. (2021), “Artificial neural
networks for educational data mining in higher education: a systematic literature review”,
Applied Artificial Intelligence, Vol. 35 No. 13, pp. 983-1021, doi: 10.1080/
08839514.2021.1922847.
Ouyang, F., Zheng, L. and Jiao, P. (2022), “Artificial intelligence in online higher education: a
systematic review of empirical research from 2011 to 2020”, Education and Information
Technologies, Vol. 27 No. 6, pp. 7893-7925, doi: 10.1007/s10639-022-10925-9.
Page, M.J., McKenzie, J.E., Bossuyt, P.M., Boutron, I., Hoffmann, T.C., Mulrow, C.D., Shamseer, L.,
Tetzlaff, J.M., Akl, E.A., Brennan, S.E. and Chou, R. (2021), “The PRISMA 2020 statement: an
updated guideline for reporting systematic reviews”, BMJ, doi: 10.1136/bmj.n71.
Pardo, A., Han, F. and Ellis, R.A. (2017), “Combining university student self-regulated learning
indicators and engagement with online learning events to predict academic performance”, IEEE
Transactions on Learning Technologies, Vol. 10 No. 1, pp. 82-92, doi: 10.1109/
TLT.2016.2639508.
Parhizkar, A., Tejeddin, G. and Khatibi, T. (2023), “Student performance prediction using datamining
classification algorithms: evaluating generalizability of models from geographical aspect”,
Education and Information Technologies, Vol. 28 No. 11, pp. 14167-14185, doi: 10.1007/
s10639-022-11560-0.
Pelima, L.R., Sukmana, Y. and Rosmansyah, Y. (2024), “Predicting university student graduation using
academic performance and machine learning: a systematic literature review”, IEEE Access,
Vol. 12, pp. 23451-23465, doi: 10.1109/ACCESS.2024.3361479.
Polyzou, A. and Karypis, G. (2019), “Feature extraction for next-term prediction of poor student
performance”, IEEE Transactions on Learning Technologies, Vol. 12 No. 2, pp. 237-248, doi:
10.1109/TLT.2019.2913358.
Qu, S., Li, K., Zhang, S. and Wang, Y. (2018), “Predicting achievement of students in smart campus”,
IEEE Access, Vol. 6, pp. 60264-60273, doi: 10.1109/ACCESS.2018.2875742.
Raj, N.S. and VG, R. (2022), “Early prediction of student engagement in virtual learning environments
using machine learning techniques”, E-learning and Digital Media, Vol. 19 No. 6, pp. 537-554,
doi: 10.1177/20427530221108027.
Ramaswami, G., Susnjak, T., Mathrani, A., Lim, J. and Garcia, P. (2019), “Using educational data
mining techniques to increase the prediction accuracy of student academic performance”,
Information and Learning Sciences, Vol. 120 Nos 7/8, pp. 451-467, doi: 10.1108/ILS-03-
2019-0017.
Riestra-Gonz�alez, M., Paule-Ru�ız, M. and Ortin, F. (2021), “Massive LMS log data analysis for the
early prediction of course-agnostic student performance”, Computers and Education, Vol. 163,
104108, doi: 10.1016/j.compedu.2020.104108.
AEDS Rodrigues, R.L., Ramos, J.L.C., Silva, J.C.S., Dourado, R.A. and Gomes, A.S. (2019), “Forecasting
students’ performance through self-regulated learning behavioral analysis”, International
Journal of Distance Education Technologies, Vol. 17 No. 3, pp. 52-74, doi: 10.4018/
IJDET.2019070104.
Sarmet, M., Kabani, A., Coelho, L., Dos Reis, S.S., Zeredo, J.L. and Mehta, A.K. (2023), “The use of
natural language processing in palliative care research: a scoping review”, Palliative Medicine,
Vol. 37 No. 2, pp. 275-290, doi: 10.1177/02692163221141969.
Sathe, M. and Adamuthe, A.C. (2021), “Comparative study of supervised algorithms for prediction of
students’ performance”, International Journal of Modern Education and Computer Science,
Vol. 13 No. 1, pp. 1-21, doi: 10.5815/ijmecs.2021.01.01.
Silva, M.P.R.I.R., Rupasingha, R.A.H.M. and Kumara, B.T.G.S. (2024), “Identifying complex causal
patterns in students’ performance using machine learning”, Technology, Pedagogy and
Education, Vol. 33 No. 1, pp. 103-119, doi: 10.1080/1475939X.2023.2288015.
So, J.C.-H., Ho, Y.H., Wong, A.K.-L., Chan, H.C.B., Tsang, K.H.-Y., Chan, A.P.-L. and Wong, S.C.-W.
(2023), “Analytic study for predictor development on student participation in generic
competence development activities based on academic performance”, IEEE Transactions on
Learning Technologies, Vol. 16 No. 5, pp. 790-803, doi: 10.1109/TLT.2023.3291310.
Son, N.T.K., Van Bien, N., Quynh, N.H. and Tho, C.C. (2022), “Machine learning based admission
data processing for early forecasting students’ learning outcomes”, International Journal of Data
Warehousing and Mining, Vol. 18 No. 1, pp. 1-15, doi: 10.4018/IJDWM.313585.
Sugden, N., Brunton, R., MacDonald, J., Yeo, M. and Hicks, B. (2021), “Evaluating student
engagement and deep learning in interactive online psychology learning activities”, Australasian
Journal of Educational Technology, Vol. 37 No. 2, pp. 45-65, doi: 10.14742/ajet.6632.
Sultana, S., Khan, S. and Abbas, M.A. (2017), “Predicting performance of electrical engineering
students using cognitive and non-cognitive features for identification of potential dropouts”,
International Journal of Electrical Engineering Education, Vol. 54 No. 2, pp. 105-118, doi:
10.1177/0020720916688484.
Trakunphutthirak, R. and Lee, V. (2022), “Application of educational data mining approach for student
academic performance prediction using progressive temporal data”, Journal of Educational
Computing Research, Vol. 60 No. 3, pp. 742-776, doi: 10.1177/07356331211048777.
Tseng, C.-H., Lin, H.-C.K., Huang, A. C.-W. and Lin, J.-R. (2023), “Personalized programming
education: using machine learning to boost learning performance based on students’ personality
traits”, Cogent Education, Vol. 10 No. 2, 2245637, doi: 10.1080/2331186X.2023.2245637.
Tsiakmaki, M., Kostopoulos, G., Kotsiantis, S. and Ragos, O. (2021), “Fuzzy-based active learning for
predicting student academic performance using autoML: a step-wise approach”, Journal of
Computing in Higher Education, Vol. 33 No. 3, pp. 635-667, doi: 10.1007/s12528-021-09279-x.
Umer, R., Susnjak, T., Mathrani, A. and Suriadi, L. (2023), “Current stance on predictive analytics in
higher education: opportunities, challenges and future directions”, Interactive Learning
Environments, Vol. 31 No. 6, pp. 3503-3528, doi: 10.1080/10494820.2021.1933542.
Veerasamy, A., Laakso, M. and D’Souza, D. (2022), “Formative assessment tasks as indicators of
student engagement for predicting at-risk students in programming courses”, Informatics in
Education, Vol. 21 No. 2, pp. 375-393, doi: 10.15388/infedu.2022.15.
Wakelam, E., Jefferies, A., Davey, N. and Sun, Y. (2020), “The potential for student performance
prediction in small cohorts with minimal available attributes”, British Journal of Educational
Technology, Vol. 51 No. 2, pp. 347-370, doi: 10.1111/bjet.12836.
Wan, H., Liu, K., Yu, Q. and Gao, X. (2019), “Pedagogical intervention practices: improving learning
engagement based on early prediction”, IEEE Transactions on Learning Technologies, Vol. 12
No. 2, pp. 278-289, doi: 10.1109/TLT.2019.2911284.
Wang, C. (2022), “Emotion recognition of college students’ online learning engagement based on deep
learning”, International Journal of Emerging Technologies in Learning (iJET), Vol. 17 No. 06,
pp. 110-122, doi: 10.3991/ijet.v17i06.30019.
Wu, J. (2021), “Learning analytics on structured and unstructured heterogeneous data sources: Asian Education
perspectives from procrastination, help-seeking, and machine-learning defined cognitive and Development
engagement”, Computers and Education, Vol. 163, 104066, doi: 10.1016/ Studies
j.compedu.2020.104066.
Xue, H. and Niu, Y. (2023), “Multi-output based hybrid integrated models for student performance
prediction”, APPLIED SCIENCES-BASEL, Vol. 13 No. 9, p. 5384, doi: 10.3390/app13095384.
Yagci, M. (2022), “Educational data mining: prediction of students’ academic performance using
machine learning algorithms”, Smart Learning Environments, Vol. 9 No. 1, p. 11, doi: 10.1186/
s40561-022-00192-z.
Y€ur€
um, O.R., Taşkaya-Temizel, T. and Yıldırım, S. (2023), “The use of video clickstream data to
predict university students’ test performance: a comprehensive educational data mining
approach”, Education and Information Technologies, Vol. 28 No. 5, pp. 5209-5240, doi:
10.1007/s10639-022-11403-y.
Zawacki-Richter, O., Mar�ın, V.I., Bond, M. and Gouverneur, F. (2019), “Systematic review of research
on artificial intelligence applications in higher education – where are the educators?”,
International Journal of Educational Technology in Higher Education, Vol. 16 No. 1, p. 39, doi:
10.1186/s41239-019-0171-0.
Zhang, W., Wang, Y. and Wang, S. (2022), “Predicting academic performance using tree-based
machine learning models: a case study of bachelor students in an engineering department in
China”, Education and Information Technologies, Vol. 27 No. 9, pp. 13051-13066, doi: 10.1007/
s10639-022-11170-w.
Zheng, X. and Li, C. (2024), “Predicting students’ academic performance through machine learning
classifiers: a study employing the Naive Bayes classifier (NBC)”, International Journal of
Advanced Computer Science and Applications, Vol. 15 No. 1, doi: 10.14569/
IJACSA.2024.0150199.

Appendix

Table A1. Algorithms used in all the studies

Methods References

DT Huang et al. (2023), Silva et al. (2024), Nti et al. (2022), Barata et al. (2016), Ayouni et al. (2021),
Wan et al. (2019), Raj and VG (2022), Veerasamy et al. (2022), Alturki and Alturki (2021), Asif
et al. (2017), Polyzou and Karypis (2019), Gray and Perkins (2019), Wakelam et al. (2020),
Zhang et al. (2022), Hussain et al. (2022), Trakunphutthirak and Lee (2022), Nachouki and Abou
Naaj (2022), Rodrigues et al. (2019), Mi et al. (2022), Durak and Bulut (2024), Parhizkar et al.
(2023), Altun et al. (2022), Xue and Niu (2023), Francis and Babu (2019), Aydo� gdu (2020), Son
et al. (2022), Mushi and Ngondya (2021), Evangelista and Sy (2022), Goundar et al. (2022), Hang
et al. (2022), Meghji et al. (2023), Sathe and Adamuthe (2021), Jamjoom et al. (2021)
RF Tseng et al. (2023), Silva et al. (2024), Nti et al. (2022), Liao and Wu (2022), Wu (2021), Raj and
VG (2022), So et al. (2023), Alturki and Alturki (2021), Asif et al. (2017), Polyzou and Karypis
(2019), Gray and Perkins (2019), Yagci (2022), Wakelam et al. (2020), Zhang et al. (2022),
Kukkar et al. (2023), Bertolini et al. (2021), Trakunphutthirak and Lee (2022), Nachouki and
Abou Naaj (2022), Alturki et al. (2022), Durak and Bulut (2024), Parhizkar et al. (2023), Xue and
Niu (2023), Liu et al. (2022), Chen and Cui (2020), Son et al. (2022), Mushi and Ngondya (2021),
Evangelista and Sy (2022), Ramaswami et al. (2019), Goundar et al. (2022), Alamgir et al.
(2024), Abdullah et al. (2023), Hang et al. (2022), Bere et al. (2022), Meghji et al. (2023), Sathe
and Adamuthe (2021)

(continued )
AEDS Table A1. Continued

Methods References

NB Barata et al. (2016), Alturki and Alturki (2021), Asif et al. (2017), Yagci (2022), Abu Zohair
(2019), Alturki et al. (2022), Mi et al. (2022), Durak and Bulut (2024), Parhizkar et al. (2023),
Zheng and Li (2024), Francis and Babu (2019), Sultana et al. (2017), Nuankaew et al. (2020),
Riestra-Gonz�alez et al. (2021), Chen and Cui (2020), Ali and Hanna (2022), Aydo� gdu (2020),
Son et al. (2022), Evangelista and Sy (2022), Ramaswami et al. (2019), Goundar et al. (2022),
Hang et al. (2022), Meghji et al. (2023), Sathe and Adamuthe (2021), Jamjoom et al. (2021)
SVM Huang et al. (2023), Silva et al. (2024), Liao and Wu (2022), Barata et al. (2016), Wu (2021)
Ayouni et al. (2021), Polyzou and Karypis (2019), Yagci (2022), Abu Zohair (2019), Hussain
et al. (2022), Rodrigues et al. (2019), Alturki et al. (2022), Mi et al. (2022), Durak and Bulut
(2024), Parhizkar et al. (2023), Francis and Babu (2019), Riestra-Gonz�alez et al. (2021), Chen
and Cui (2020), Ali and Hanna (2022), Mushi and Ngondya (2021), Evangelista and Sy (2022),
Goundar et al. (2022), Meghji et al. (2023), Sathe and Adamuthe (2021), Jamjoom et al. (2021)
KNN Huang et al. (2023), Barata et al. (2016), Raj and VG (2022), So et al. (2023), Asif et al. (2017),
Gray and Perkins (2019), Yagci (2022), Abu Zohair (2019), Wakelam et al. (2020),
Trakunphutthirak and Lee (2022), Alturki et al. (2022), Durak and Bulut (2024), Chen and Cui
(2020), Alharbi et al. (2021), Aydo� gdu (2020), Mushi and Ngondya (2021), Evangelista and Sy
(2022), Ramaswami et al. (2019), Bere et al. (2022), Sathe and Adamuthe (2021), Jamjoom et al.
(2021)
LGR Huang et al. (2023), Barata et al. (2016), Raj and VG (2022), Yagci (2022), Bertolini et al. (2021),
Trakunphutthirak and Lee (2022), Rodrigues et al. (2019), Alturki et al. (2022), Mi et al. (2022),
Durak and Bulut (2024), Altun et al. (2022), Sultana et al. (2017), Riestra-Gonz�alez et al. (2021),
Chen and Cui(2020), Son et al. (2022), Evangelista and Sy (2022), Ramaswami et al. (2019),
Hang et al. (2022)
ANN Wu (2021), Ayouni et al. (2021), So et al. (2023), Asif et al. (2017), Abu Zohair (2019), Alturki
et al. (2022), Altun et al. (2022), Nuankaew et al. (2020), Aydo� gdu (2020), Hang et al. (2022)
GB Polyzou and Karypis (2019), Zhang et al. (2022), Kukkar et al. (2023), Bertolini et al. (2021),
Trakunphutthirak and Lee (2022), Xue and Niu (2023), Liu et al. (2022), Y€ ur€um et al. (2023),
Chen and Cui (2020), Abdullah et al. (2023)
NN Trakunphutthirak and Lee (2022), Francis and Babu (2019), Liu (2019), Sultana et al. (2017),
Chen and Cui (2020), Alamgir et al. (2024)
MLP Silva et al. (2024), Riestra-Gonz�alez et al. (2021), Mushi and Ngondya (2021), Evangelista and
Sy (2022), Bere et al. (2022)
LSTM Liao and Wu (2022), Wang (2022), Kukkar et al. (2023), Liu et al. (2022), Chen and Cui (2020)
K-Means Moubayed et al. (2021), Nkomo and Nat (2021), De Backer et al. (2022a, b), Francis and Babu
(2019) Ali and Hanna (2022)
AdaBoost Hussain et al. (2019), Xue and Niu (2023), Abdullah et al. (2023)
LR So et al. (2023), Altun et al. (2022), Aydo� gdu (2020), Alamgir et al. (2024)
CNN Alshraideh et al. (2024), Wang (2022), Parhizkar et al. (2023)
HC De Backer et al. (2022a, b), Pardo et al. (2017), Ali and Hanna (2022)
LADTree Alturki and Alturki (2021), Abu Zohair (2019), Son et al. (2022)
Jrip Hussain et al. (2019), Murata et al. (2023)
RNN Mi et al. (2022), Hang et al. (2022)
DNN Liao and Wu (2022), Liu et al. (2022)
FNN Evangelista and Sy (2022)
OneR Evangelista and Sy (2022), Meghji et al. (2023)
NL Yagci (2022)
ENR Bertolini et al. (2021)
BR Abdullah et al. (2023)
SVD Alharbi et al. (2021)
NMF Alharbi et al. (2021)
BA Alharbi et al. (2021)
SA Alharbi et al. (2021)
FCM Tsiakmaki et al. (2021)
Kstar Meghji et al. (2023)
Robust Y€ um et al. (2023)
ur€

(continued )
Table A1. Continued Asian Education
and Development
Methods References Studies
Voting Aggarwal et al. (2021)
Bagging Aggarwal et al. (2021)
LightGBM Xue and Niu (2023)
BN Alturki and Alturki (2021)
K- Sugden et al. (2021)
Medoids
Stacking So et al. (2023)
EM Ali and Hanna (2022)
ILDA Ali and Hanna (2022)
CA Alharbi et al. (2021)
Source(s): Authors’ own creation

Table A2. Evaluation methodologies in all studies

Evaluation
metrics References

AC Huang et al. (2023), Silva et al. (2024), Barata et al. (2016), Wu (2021), Wang (2022), Raj
and VG (2022), Alturki and Alturki (2021), Asif et al. (2017), Gray and Perkins (2019),
Yagci (2022), Abu Zohair (2019), Hussain et al. (2019), Hussain et al. (2022), Kukkar et al.
(2023), Tsiakmaki et al. (2021), Bertolini et al. (2021), Trakunphutthirak and Lee (2022),
Rodrigues et al. (2019), Alturki et al. (2022), Mi et al. (2022), Durak and Bulut (2024),
Parhizkar et al. (2023), Zheng and Li (2024), Xue and Niu (2023), Francis and Babu (2019),
Nuankaew et al. (2020), Riestra-Gonz�alez et al. (2021), Ali and Hanna (2022), Aydo� gdu
(2020), Mushi and Ngondya (2021), Evangelista and Sy (2022), Goundar et al. (2022),
Hang et al. (2022), Bere et al. (2022), Meghji et al. (2023), Sathe and Adamuthe (2021),
Jamjoom et al. (2021)
F1 Huang et al. (2023), Silva et al. (2024), Wu (2021), Wang (2022), Alturki and Alturki
(2021), Polyzou and Karypis (2019), Gray and Perkins (2019), Yagci (2022), Hussain et al.
(2019), Zhang et al. (2022), Hussain et al. (2022), Kukkar et al. (2023), Tsiakmaki et al.
(2021), Trakunphutthirak and Lee (2022), Alturki et al. (2022), Mi et al. (2022), Parhizkar
et al. (2023), Zheng and Li (2024), Xue and Niu (2023), Francis and Babu (2019), Liu et al.
(2022), Murata et al. (2023), Riestra-Gonz�alez et al. (2021), Ali and Hanna (2022), Mushi
and Ngondya (2021), Evangelista and Sy (2022), Alamgir et al. (2024), Bere et al. (2022),
Meghji et al. (2023), Jamjoom et al. (2021)
Recall Huang et al. (2023), Silva et al. (2024), Wu (2021), Wang (2022), Raj and VG (2022),
Veerasamy et al. (2022), Yagci (2022), Hussain et al. (2019), Zhang et al. (2022), Hussain
et al. (2022), Kukkar et al. (2023), Trakunphutthirak and Lee (2022), Rodrigues et al.
(2019), Alturki et al. (2022), Mi et al. (2022), Durak and Bulut (2024), Parhizkar et al.
(2023), Zheng and Li (2024), Xue and Niu (2023), Francis and Babu (2019), Murata et al.
(2023), Nuankaew et al. (2020), Ali and Hanna (2022), Mushi and Ngondya (2021),
Evangelista and Sy (2022), Goundar et al. (2022), Hang et al. (2022), Sathe and Adamuthe
(2021), Jamjoom et al. (2021)

(continued )
AEDS Table A2. Continued

Evaluation
metrics References

Precision Huang et al. (2023), Silva et al. (2024), Wu (2021), Wang (2022), Raj and VG (2022),
Polyzou and Karypis (2019), Yagci (2022), Hussain et al. (2019), Zhang et al. (2022),
Hussain et al. (2022), Kukkar et al. (2023), Trakunphutthirak and Lee (2022), Rodrigues
et al. (2019), Alturki et al. (2022), Durak and Bulut (2024) , Parhizkar et al. (2023), Zheng
and Li (2024), Xue and Niu (2023), Francis and Babu (2019), Murata et al. (2023),
Nuankaew et al. (2020), Ali and Hanna (2022), Mushi and Ngondya (2021), Evangelista
and Sy (2022), Bere et al. (2022), Sathe and Adamuthe (2021), Jamjoom et al. (2021)
AUC Huang et al. (2023), Liao and Wu (2022), Wan et al. (2019), Alturki and Alturki (2021),
Polyzou and Karypis (2019), Gray and Perkins (2019), Yagci (2022), Tsiakmaki et al.
(2021), Rodrigues et al. (2019), Alturki et al. (2022), Liu et al. (2022), Riestra-Gonz�alez
et al. (2021), Chen and Cui (2020), Alamgir et al. (2024), Hang et al. (2022), Jamjoom et al.
(2021)
CV Liao and Wu (2022), Alshraideh et al. (2024), Barata et al. (2016), Wu (2021), So et al.
(2023), Yagci (2022), Abu Zohair (2019), Tsiakmaki et al. (2021), Trakunphutthirak and
Lee (2022), Nachouki and Abou Naaj (2022), Durak and Bulut (2024), Liu et al. (2022),
Y€ur€um et al. (2023), Mushi and Ngondya (2021), Evangelista and Sy (2022)
MAE Silva et al. (2024), Nachouki and Abou Naaj (2022), Altun et al. (2022), Xue and Niu
(2023), Y€ um et al. (2023), Alharbi et al. (2021), Alamgir et al. (2024), Abdullah et al.
ur€
(2023)
RMSE Silva et al. (2024), Nti et al. (2022), So et al. (2023), Nachouki and Abou Naaj (2022),
Altun et al. (2022), Y€ um et al. (2023), Alharbi et al. (2021), Son et al. (2022), Alamgir
ur€
et al. (2024), Abdullah et al. (2023)
CMT Veerasamy et al. (2022), Asif et al. (2017), Yagci (2022), Sultana et al. (2017), Nuankaew
et al. (2020), Ramaswami et al. (2019), Sathe and Adamuthe (2021)
Kappa Asif et al. (2017), Abu Zohair (2019), Hussain et al. (2019), Goundar et al. (2022), Hang
et al. (2022), Meghji et al. (2023)
MSE Wakelam et al. (2020), Nachouki and Abou Naaj (2022), Altun et al. (2022), Xue and Niu
(2023)
MAPE Nti et al. (2022), Wakelam et al. (2020), Altun et al. (2022), Abdullah et al. (2023)
SP Wu (2021), Veerasamy et al. (2022), Goundar et al. (2022), Hang et al. (2022)
R2 Altun et al. (2022), Xue and Niu (2023), Abdullah et al. (2023)
SD Y€ur€um et al. (2023), Riestra-Gonz�alez et al. (2021)
JHO Y€ur€um et al. (2023), Ramaswami et al. (2019)
CC Nti et al. (2022), Wakelam et al. (2020)
KR20 Liao and Wu (2022), Wu (2021)
OOB Tseng et al. (2023), Rodrigues et al. (2019)
AN De Backer et al. (2022a, b), Bertolini et al. (2021)
95% CI Riestra-Gonz�alez et al. (2021)
PTA Alharbi et al. (2021)
SC Moubayed et al. (2021)
MWU Tseng et al. (2023)
T Moubayed et al. (2021)
α Wu (2021)
MA De Backer et al. (2022a, b)
JC Bertolini et al. (2021)
SI Abdullah et al. (2023)
Pearson Abdullah et al. (2023)
Spearman Abdullah et al. (2023)
Source(s): Authors’ own creation
Table A3. Results of the application of machine learning algorithms in higher education Asian Education
and Development
No Authors Findings Studies
1 Huang et al. (2023) It was found that the individualized intervention not only improved
students’ academic performance but also promoted students’ competence
in the following learning strategies: rehearsal, critical thinking,
metacognitive self-regulation, effort regulation and peer learning.
Additionally, students’ rehearsal and help-seeking learning strategies
indirectly impacted academic performance through students’ notes in the
provided e-book
2 Tseng et al. (2023) It was found that GSR and heart rate variability significantly predicted
extraversion, while heart rate variability also predicted affinity and
conscientiousness. In the case study, students who used personalized
learning materials performed significantly better than other students
3 Silva et al. (2024) It was found that among the separate classification algorithms, the MLP
algorithm had the highest accuracy and lowest error value. In addition,
when the five algorithms were combined using an integrated learning
approach, better results were obtained, showing that integrated learning
improves accuracy
4 Nti et al. (2022) It was found that the frequency and nature of social media use in the
classroom partially affects students’ academic performance. In particular,
the use of social media within the classroom (multitasking), had a
significant negative impact on students’ GPA
5 Liao and Wu (2022) It was found that students who reported more digitally disruptive problems
received lower final course grades, while students who reported stronger
peer learning orientations received higher final course grades.
Additionally, the predictive validity of peer learning involvement on
academic performance as objectively identified by the ML model was
superior to that of self-perceived peer learning orientation
6 Alshraideh et al. (2024) The study found that 96.69 percent accuracy could be achieved in
predicting a student’s GPA through a CNN model. The study also found
that if students continued to play online games every day, 80 percent of the
students’ GPAs decreased, 2 percent of the students’ GPAs increased, and
the rest of the students remained unchanged
7 Barata et al. (2016) The study found that using performance data alone, it was possible to
predict student type at mid-course with 79 percent accuracy. When
combined with game data, the accuracy of the predictions improved in the
early stages of the programme
8 Moubayed et al. (2021) It was found that the number of logins and the average duration of
submitted assignments were representative indicators of student
engagement, while the number of content reads was not. Additionally, the
two-level model performed best in terms of cluster separation, but the
three-level model performed similarly while better identifying low-
engaged students
9 Sugden et al. (2021) The study found that students commonly use multiple devices to learn in
different locations and that these online activities are perceived to
contribute to deep learning and engagement. Students had positive
perceptions of the activities, particularly those that were authentic,
promoted problem solving and applied theory to real-life situations
10 Wu (2021) It was found that help-seeking behaviors positively predicted statistical
achievement, while ICT-related academic procrastination negatively
predicted statistical achievement. In addition, messages categorized by
integrated machine learning were highly consistent with human-coded
messages in terms of the degree of correlation and were positively
associated with statistical achievement

(continued )
AEDS Table A3. Continued

No Authors Findings

11 Wang (2022) It is found that the proposed model performs well in emotion recognition,
with a stable emotion recognition accuracy of approximately 96% on both
the training and validation sets, and a loss rate of approximately 0.06. In
addition, the proposed model performs best in emotion recognition
compared to the other five models including CNN, LSTM, BLSTM,
Attention-based BLSTM and EWW
12 Ayouni et al. (2021) The study found that the machine learning algorithm was able to predict the
level of student engagement. ANN outperformed SVM (80%), and DT
(75%), in terms of accuracy (85%), recall and F1 score
13 Wan et al. (2019) It was found that learner engagement can be significantly increased by
reminder interventions, and error-related recommendation interventions
can also play a positive role in increasing learning engagement
14 Raj and VG (2022) It was found that the Random Forest classification algorithm performed
best in predicting student engagement with 95% accuracy, 95% precision,
and 98% correlation
15 Veerasamy et al. (2022) It was found that ongoing assessment was found to correlate with student
engagement and performance on the final programming exam, and student
performance could be predicted with considerable accuracy (RMSE
percentages between 15 and 20 per cent), from ongoing assessment data
16 So et al. (2023) It was found that there was a significant difference between students’ DSE
P scores and the amount and duration of their participation in GCDA,
indicating that students with stronger academic backgrounds were more
inclined to participate in GCDA. In addition, there was a positive
correlation between the level of GCDA participation and students’ CGPA
17 Nkomo and Nat (2021) The study found different patterns of student interaction with learning
resources based on their level of engagement (low, medium, and high).
Students with high levels of engagement visited course pages more
frequently, participated in forum discussions, and achieved higher grades
in the course
18 De Backer et al. (2022a, b) The study found comprehensive affirmative regulators (AOAR), socially
oriented exhaustive regulators (SOER), and individually oriented passive
regulators (IOPR). These types showed significant differences in
conceptual understanding, motivation to learn, and self-efficacy
19 Alturki and Alturki(2021) It was found that a student’s GPA during the first four semesters, the
number of courses failed within the first two semesters, and grades in the
three core courses (Database Fundamentals, Programming Languages, and
Computer Network Fundamentals), were the main characteristics that
predicted student academic achievement. In addition, the Naı€ve Bayes
classifier outperformed the tree-based model in predicting student
academic achievement overall, but the Random Forest performed better in
predicting honors students
20 Asif et al. (2017) Research has found that by focusing on a small number of courses, it is
possible to provide timely early warning and support for underperforming
students, as well as advice and opportunities for high-performing students
21 Polyzou and Karypis It was found that the Gradient Boosting (GB), classifier performed best in
(2019) all classification tasks, with its maximum F1 score of 0.118 in the task of
predicting student failure. Student’s historical performance characteristics
were highly predictive in predicting student performance, especially in the
tasks of predicting student withdrawal and relative underperformance
22 Pardo et al. (2017) Self-efficacy and intrinsic motivation were found to be positively
associated with positive self-regulation strategy use, while test anxiety and
negative self-regulation strategy use were negatively associated with
academic performance. Combining self-regulated learning variables and
participation in online learning events explained 32% of the variance in
academic performance

(continued )
Table A3. Continued Asian Education
and Development
No Authors Findings Studies
23 Gray and Perkins (2019) The study found that students could be identified with approximately 97
percent accuracy using Bangor Engagement Metric (BEM), values from
just the first three weeks of school
24 Yagci (2022) It was found that the proposed model achieved 70-75% classification
accuracy. In particular, Random Forest, Neural Network and Support
Vector Machine algorithms performed best in predicting students’
academic performance
25 Abu Zohair (2019) It was found that Support Vector Machine (SVM), and Linear Discriminant
Analysis (LDA), algorithms are very effective in training and producing
acceptable classification accuracy and reliability test rates on small
datasets.
26 Wakelam et al. (2020) The study found that predictions using a limited number of student
attributes are feasible in small student populations, and that these
predictions are useful in supporting course leaders in identifying
potentially “at-risk” students. Prediction accuracy improved as the course
progressed, with Random Forest and K-Nearest Neighbor algorithms
producing the most accurate predictions
27 Hussain et al. (2019) The study found that deep learning techniques achieved a classification
accuracy of 95.34 percent in predicting student academic performance,
which was the highest of all the methods compared
28 Zhang et al. (2022) The study found that the Random Forest (RF), model was successful in
identifying over 80 percent of low performance risk students by the end of
the second semester
29 Hussain et al. (2022) It was found that the SVM algorithm was more accurate than the decision
tree algorithm for a dataset containing 260 records, while the decision tree
algorithm was slightly more accurate than the SVM algorithm for a dataset
containing 520 records. Both algorithms have the same accuracy of
74.04% when the data split is 80%-20%
30 Kukkar et al. (2023) It was found that the proposed SAPP system achieved approximately 96
percent in prediction accuracy, which is more than that of the existing
systems. The inclusion of affective information improves the accuracy of
the prediction model
31 Tsiakmaki et al. (2021) It is found that tree-based classifiers (e.g. PTs and PTTDs), outperform
other fuzzy classifiers in most cases. Active learning strategies are able to
construct efficient predictive models while reducing the amount of training
data
32 Bertolini et al. (2021) It was found that feature selection techniques CAE and FSA significantly
improved the prediction of LR and GLMNET models. Whereas RAE
algorithm performed poorly in terms of stability and predictive
performance. Borda method identified GPA, number of credits taken and
performance on concept inventory assessment as the major factors
affecting the prediction of student performance
33 Trakunphutthirak and Lee It was found that a combination of internet usage log files and demographic
(2022) data could improve prediction accuracy. In particular, random forest and
decision tree algorithms performed well in predicting academic
performance
34 Nachouki and Abou Naaj The study found that CPM achieved accurate results in predicting student
(2022) performance, with higher accuracy compared to the benchmark method
35 Rodrigues et al. (2019) It was found that the logistic regression model performed best in predicting
students’ academic performance with an accuracy of 0.893 and an area
under the ROC curve of 0. 9574. This means that the logistic regression
model was effective in differentiating and predicting students’ academic
success

(continued )
AEDS Table A3. Continued

No Authors Findings

36 Alturki et al. (2022) The study found that semester grades were the most important
characteristic in predicting students’ academic achievement, followed by
distance between the student’s residential location and the university and
cultural background. In addition, gender, age, number of failed courses,
and number of registered and unregistered exams per semester had less of
an impact on the predictions
37 Mi et al. (2022) It is found that the proposed model achieves an F1 score of 99.25% on
large-scale datasets, which is 1.25% higher than other relevant models
38 Durak and Bulut (2024) The study found that the means of students’ programming performance,
computational identity, computational thinking perspectives, and
programming empowerment were significantly different on the gender and
education level variables. However, the study variables were not
significantly different on the academic achievement level variable
39 Parhizkar et al. (2023) It was found that with careful feature selection, data collection and model
training, generalized models can be trained that consider geographical
differences. The experimental results show that the best model trained
using a specific dataset has better generalization ability compared to the
base model trained and tested only on Iranian domestic data
40 Altun et al. (2022) The study found that the developed model was able to predict students’
graduation grades from first semester data with 94 to 97 percent accuracy.
In addition, the early warning model was able to predict with 72 to 87
percent accuracy whether a student’s weighted GPAwould fall below 2 in a
future semester
41 Zheng and Li (2024) It was found that in the prediction of G1, the F1 score of the NBAR model
was 0.882, which was approximately 1.03 percent higher than that of the
NBJS model of 0.873. In G3 prediction, the F1 score of the NBAR model
was 0.893 compared to 0.884 for the NBJS model
42 Xue and Niu (2023) It was found that the accuracy of predicting midterm and final grades using
the XGBoost model and the mean interpolation method (mean imputation),
was 78.37%, which was 3-8% better than the comparison model. Using the
Gdbt model to predict homework and lab grades, the average root mean
square error was 16.76, which was better than the comparison model
43 Aggarwal et al. (2021) It was found that when both academic and nonacademic parameters were
used, the F1 scores of the model were significantly higher, indicating the
importance of nonacademic parameters in predicting student performance
44 Francis and Babu (2019) It was found that a hybrid algorithm combining clustering and
classification methods had high accuracy in predicting student academic
performance
45 Liu, 2019 It was found that NCEE scores had the greatest impact on college students’
English performance, followed by learning attitude and gender, while age
had little effect on English performance. The improved BPNN achieved
good results in predicting college students’ English performance
46 Liu et al. (2022) The study found that the LSTM algorithm performed best on all
assessment metrics with 90.25% accuracy. In addition, the study found that
the four learning sites (content, subpages, homepage, and quizzes), in the
clickstream data had a significant impact on student course performance
47 Y€ um et al. (2023)
ur€ A positive correlation was found between student test performance and
total number of clicks, slow backing, and performing backing or pausing.
In particular, the mean of backing speed, the number of pauses, and the
number/percentage of backing were found to be the most important
predictors of student test performance
48 Sultana et al. (2017) It was found that the prediction accuracy including noncognitive features
improved when the decision tree method was used. However, logistic
regression, Naı€ve Bayes(NB), and neural networks showed a decrease in
prediction accuracy when including noncognitive features

(continued )
Table A3. Continued Asian Education
and Development
No Authors Findings Studies
49 Murata et al. (2023) It was found that RNN-FitNets improve the accuracy of early predictions
and that, through SHAP value analysis, RNN-FitNets are able to take into
account the future effects of inputting features from the early stages of a
course
50 Nuankaew et al. (2020) It was found that the model developed using the artificial neural network
algorithm had the highest accuracy in the first three datasets (89.04%,
92.70% and 93.71% accuracy, respectively), while the last dataset (91.68%
accuracy), was suitable for the use of the Naı€ve Bayes(NB)ian algorithm
51 Riestra-Gonz�alez et al. It was found that the accuracy of all models improved as the prediction
(2021) moment progressed. MLP performed best in all scenarios, improving from
80.1% accuracy at 10% course completion to 90.1% at 50%
52 Chen and Cui (2020) Deep learning methods using LSTM networks were found to have higher
accuracy and generalization capabilities than traditional machine learning
methods in predicting student performance. Click frequency based on
time-series information was effective in early detection of at-risk students
53 Ali and Hanna (2022) The study found that students in the high SRL group achieved higher
grades than students in the medium and low SRL groups. Disaggregated
results showed that log data and engagement activities successfully
predicted students’ academic performance with over 88% accuracy
54 Alharbi et al. (2021) It was found that the K-Nearest Neighbor and ZScore models performed
best in predicting student achievement, particularly achieving 84%
accuracy on PTA2 (the percentage of differences between predicted letter
grades and actual letter grades that do not exceed two consecutive letter
grades)
55 Aydo�
gdu (2020) It was found that 80.47% prediction accuracy could be achieved by tuning
the parameters of the ANN model. Specifically, the model performed best
when there were eight neurons in the hidden layer, a batch size of 128, a few
iterations of 1,000, and when the Adam optimization algorithm was used.
Additionally, the study found that the number of live sessions attended, the
number of archived sessions attended, and the time spent on content were
the three most significant factors influencing student performance
56 Son et al. (2022) It was found that the accuracy of predictive models can be improved by
selecting important attributes and removing redundant and irrelevant
information from the dataset. In addition, it was found that Mathematics
was the most important factor influencing the output results of the subject
combination of the entrance examination for students of elementary
education at Hanoi Metropolitan University
57 Mushi and Ngondya It was found that the RF algorithm performed best in predicting math
(2021) performance with 99% accuracy and achieved F1 scores of 99% and 100%
for failing and passing classifications, respectively
58 Evangelista and Sy (2022) It was found that the Bagging integration method using the OneR
algorithm as the base classifier performed the best in terms of prediction
accuracy, with an increase in accuracy from 93.10 percent to 93.62 percent.
In addition, the heterogeneous integration method (voting), slightly
outperformed the homogeneous integration methods (Bagging and
Boosting)
59 Ramaswami et al. (2019) It was found that when general features extracted from student activities
were combined with process mining features, the accuracy of predictions
could be improved. Of the four algorithms, Random Forest was statistically
significantly more accurate than the other three methods
60 Goundar et al. (2022) The study found that the Random Forest algorithm performed best in
predicting student performance with 93.67 percent accuracy. In addition,
the study found that data from student interactions early in the course
predicted their eventual course performance

(continued )
AEDS Table A3. Continued

No Authors Findings

61 Alamgir et al. (2024) The study found that a student’s GPA in the early semesters was a good
predictor of future performance. In addition, additional characteristics
including student grades in core computer science courses and the number
of repeated courses significantly improved prediction accuracy. Random
forests performed best in most cases
62 Abdullah et al. (2023) The study found that the integrated model of Random Forest and XGBoost
performed the best. In addition, the study found that among all the
e-learning components and events, the quiz event had a significant impact
on the prediction of students’ academic performance. The study also
showed that the events in weeks 9 to 12 of the semester had an impact on
student performance throughout the semester
63 Hang et al. (2022) Multiscenario behavioral performance metrics were found to have strong
predictive power, with deep neural network models having the highest
predictive accuracy (82%), but the longest computational time. Rule-set-
based models were highly accurate, readable and actionable, facilitating
accurate instructional interventions and resource recommendations
64 Bere et al. (2022) It was found that the kNN algorithm performed best in terms of accuracy,
precision and F1 score with 93% accuracy, 90% precision and 73% F1
score
65 Meghji et al. (2023) It was found that the use of feature-selected subsets of academic and
derived attributes successfully predicted student performance at the end of
the degree. In addition, the model generated by the J48 classifier was used
to identify the courses that influence students’ final performance
66 Sathe and Adamuthe It was found that Random Forest and C5.0 outperform J48, CART, NB,
(2021) KNN and SVM on all datasets
67 Jamjoom et al. (2021) It was found that the Decision Tree and Support Vector Machine models
performed the best in predicting performance with an accuracy of 99.18
percent. It was also found that there was a strong positive correlation
between the actual assessment associated with the course and the students’
ability to pass the course
Source(s): Authors’ own creation

Corresponding author
Xiaodeng Zhou can be contacted at: [email protected]

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