Students’ Performance in Learning Management System: An Approach
to Key Attributes Identification and Predictive Algorithm Design
Dynil Duch
1,2 a
, Madeth May
1 b
and S
´
ebastien George
1 c
1
LIUM, Le Mans Universit
´
e, 72085 Le Mans, Cedex 9, France
2
Institute of Digital Research & Innovation, Cambodia Academy of Digital Technology, Phnom Penh, Cambodia
Keywords:
Deep Analytics, Knowledge Extraction, Student Performance Prediction, Moodle, Random Forest Classifier,
Predictive Algorithm, Data Normalization, Student Engagement Activities, Student Achievement.
Abstract:
The study we present in this paper explores the use of learning analytics to predict students’ performance in
Moodle, an online Learning Management System (LMS). The student performance, in our research context,
refers to the measurable outcomes of a student’s academic progress and achievement. Our research effort aims
to help teachers spot and solve problems early on to increase student productivity and success rates. To achieve
this main goal, our study first conducts a literature review to identify a broad range of attributes for predicting
students’ performance. Then, based on the identified attributes, we use an authentic learning situation, lasting
a year, involving 160 students from CADT (Cambodia Academy of Digital Technology), to collect and analyze
data from student engagement activities in Moodle. The collected data include attendance, interaction logs,
submitted quizzes, undertaken tasks, assignments, time spent on courses, and the outcome score. The collected
data is then used to train with different classifiers, thus allowing us to determine the Random Forest classifier
as the most effective in predicting students’ outcomes. We also propose a predictive algorithm that utilizes
the coefficient values from the classifier to make predictions about students’ performance. Finally, to assess
the efficiency of our algorithm, we analyze the correlation between previously identified attributes and their
impact on the prediction accuracy.
1 INTRODUCTION
In recent years, the use of Learning Management
Systems (LMS) has grown in popularity primarily
for their ability to manage and organize digital in-
formation related to teaching and learning. Educa-
tional institutions use LMS platforms such as Moo-
dle, an open-source system widely adopted in the
education sector, to facilitate the creation, distribu-
tion, and management of online learning materials
(Costaa et al. (2015)). In addition, to support teaching
practices, various studies have employed data min-
ing techniques to predict students’ performance (SP)
in Moodle and other LMS platforms, such as F
´
elix
et al. (2018). These studies have demonstrated that
making use of data from a learning environment and
efficient predictive techniques can assist teachers in
evaluating SP. For instance, a teacher can identify
a
https://orcid.org/0000-0002-7857-5811
b
https://orcid.org/0000-0002-8527-7345
c
https://orcid.org/0000-0003-0812-0712
areas where students may be struggling, and imple-
ment targeted interventions to improve student out-
comes. Therefore, the ability to predict SP can sig-
nificantly impact learning practices in general, as it
allows for more personalized and adaptive learning
experiences that can better meet the needs of indi-
vidual students. However, it is challenging to iden-
tify the crucial data attributes (hereafter referred to as
”key attributes” or ”attributes”) from online learning
activities because of the large amount of information
in the LMS (Venkatachalam et al. (2011)), especially
when the goal is to obtain a more accurate SP predic-
tion (Albreiki et al. (2021)). Moreover, predicting SP
and its associated issues have always been a consid-
erable concern in education (Ramesh et al. (2013)).
To state an example, both the nature and quality of
data strongly impact the efficiency of a predictive ap-
proach. Then, the outcomes of using the predictive
techniques heavily rely on teachers’ technical skills.
The research question in this study is: How can
teachers employ predictive techniques to evaluate SP
effectively? This question aims to explore the poten-
Duch, D., May, M. and George, S.
Students’ Performance in Learning Management System: An Approach to Key Attributes Identification and Predictive Algorithm Design.
DOI: 10.5220/0012754200003756
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 13th International Conference on Data Science, Technology and Applications (DATA 2024), pages 285-292
ISBN: 978-989-758-707-8; ISSN: 2184-285X
Proceedings Copyright © 2024 by SCITEPRESS – Science and Technology Publications, Lda.
285

tial benefits and limitations of using predictive analyt-
ics in the educational context and identify best prac-
tices for implementing these techniques to support
student learning. By addressing this research ques-
tion, educators and researchers can better understand
the concrete applications of predictive analytics in the
classroom, ultimately contributing to more effective
teaching practices and improved student outcomes.
In the context of this research, SP refers to
the measurable outcomes of a student’s academic
progress and achievement, particularly in relation to
their use of an online LMS. It includes factors such as
grades, test scores, completion of online activities and
assignments, engagement with digital learning mate-
rials and resources within the LMS. It is essential to
distinguish academic performance from performance
in a broader sense. For example, while academic
performance focuses specifically on a student’s suc-
cess in school or university, performance in a broader
sense could refer to a range of activities or tasks in
various settings. It is also crucial to note that SP is
not solely focused on positive outcomes. In addition
to measuring success, SP can also be used to iden-
tify specific learning activities or overall experience
in using the LMS so that teachers can take appropri-
ate actions in both assistance and evaluation tasks.
To address this research question, we have two
main objectives:
1. Identify the key attributes in LMS data that are rel-
evant for predicting students’ performance. Many
researchers have identified the specific attributes
for predicting SP as presented later in section 2.
However, there is a need for more identification
of various attributes that could help educators or
teachers with the selection process. Our research
will conduct a literature review by analyzing var-
ious data points, such as grades, assessment re-
sults, engagement metrics, demographic informa-
tion, and teacher pedagogies, to determine which
factors could be used for predicting SP.
2. Propose an appropriate prediction algorithm to
forecast SP based on the critical attributes identi-
fied in the first objective. This includes using pre-
dictive algorithms and data mining techniques to
analyze the data and identify patterns and trends
that can be utilized to predict future SP.
To test our research question, we have developed
the following hypotheses:
1. Hypothesis 1: there is a statistically significant
relationship between student attendance (mea-
sured by the number of modules completed) and
academic performance (measured by the final
grades) in Moodle. This hypothesis investigates
how consistent attendance influences SP predic-
tion.
2. Hypothesis 2: the extent of student interaction
with the LMS (measured by the number of inter-
action logs) has a statistically significant impact
on academic performance (measured by the final
grades). This hypothesis aims to explore whether
higher levels of engagement with the LMS posi-
tively influence SP.
3. Hypothesis 3: the number of quizzes and tasks
submitted by students statistically impacts their
final academic performance. This hypothesis ex-
amines whether students’ active participation in
quizzes and tasks within the LMS significantly
predicts their overall performance.
We use data from the Cambodia Academy of Dig-
ital Technology (CADT) to conduct our first tests,
as detailed in section 3. It is worth mentioning that
the access and the use of students’ data by all re-
search partners from France are regulated. Our re-
search complies with both Cambodian local regula-
tions and European GDPR. All personal and sensitive
information has been stripped of all data samples we
use in our research.
The rest of the paper presents our approach to
identifying a wide range of key attributes and ap-
propriate prediction algorithms. Our hypotheses will
help us better understand the impact of attributes we
have collected from CADT. The second section pro-
vides an overview of the most relevant related works.
We examine critical attributes for predicting SP and
explore the technology context and methodology ap-
proach in sections 3. We discuss our experimental
results in section 4 and conclude this paper by high-
lighting significant areas that we have been working
on since completing the tests we have conducted with
our first three hypotheses.
2 RELATED WORK
This section is dedicated to a comprehensive review
of current research on predicting student outcomes
using machine learning and data mining techniques
in educational settings that exploit LMS data, such
as Moodle. The studies we mentioned here explore
various aspects of SP, such as academic performance,
retention, and learning behaviors. The studies also
cover different techniques, including Decision Trees,
Neural Networks, Logistic Regression, Support Vec-
tor Machines, Na
¨
ıve Bayes, and Random Forests.
A meta-study by F
´
elix et al. (2018); Namoun and
Alshanqiti (2020) systematically reviewed papers that
DATA 2024 - 13th International Conference on Data Science, Technology and Applications
286

used data mining and machine learning to predict stu-
dent outcomes as a proxy for student SP. The re-
view stated that existing studies mainly focused on the
course level, using predictors such as previous aca-
demic performance, demographic data, and course-
related variables. Machine learning algorithms, in-
cluding Decision Trees, Neural Networks, Support
Vector Machines, Na
¨
ıve Bayes, and Random Forests,
were found to accurately predict student outcomes,
with some studies reporting prediction accuracies of
over 90%. However, the review also noted limitations,
such as the lack of validation of the models on new
datasets, limited explanatory power, lack of standard-
ized evaluation metrics, and potential ethical concerns
related to using sensitive student data.
In another study, Felix et al. (2019) used a dataset
of 1,307 students’ activity logs in a course, includ-
ing variables related to student interactions in forums,
chats, quizzes, activities, time spent on the platform,
and grades. They built a predictive model of student
outcomes using Na
¨
ıve Bayes, Decision Trees, Mul-
tilayer Perceptron, and Regression algorithms, with
the Na
¨
ıve Bayes model performing the best with an
accuracy of 87%. The study found that the number
of interactions with the system, attendance, and time
spent on the platform were essential variables in pre-
dicting student outcomes. Nevertheless, the study was
limited to a single course and did not consider other
factors influencing student outcomes, such as prior
knowledge or motivation.
The study of Hirokawa (2018) used machine
learning methods like Random Forests, Support Vec-
tor Machines, and Decision Trees to forecast aca-
demic achievement. The study unveiled that previous
academic records were essential for predicting aca-
demic performance, followed by the student’s gender
and age. At the same time, other attributes, such as
extracurricular activities and family background, had
a lesser impact. Yet, the study showed some limita-
tions, as it did not focus on data from LMS activities
and excluded influence factors such as teacher peda-
gogies.
In the same context, Gaftandzhieva et al. (2022)
used a machine learning algorithm to predict stu-
dents’ final grades in an Object-Oriented Program-
ming course using data from Moodle LMS activities.
The study found that the Random Forest algorithm
had the highest prediction accuracy of 78%, and at-
tendance was strongly correlated with final grades.
The study’s weaknesses, however, included its limited
sample size and singular course focus. Other stud-
ies have used data mining and machine learning to
predict the likelihood of students dropping out of a
course (Quinn and Gray (2019)), the likelihood of stu-
dent’s success in a course (Arizmendi et al. (2022)),
or predicting student grades using both academic and
non-academic factors (Ya
˘
gcı (2022)). Some studies
have also focused on predicting student outcomes in
specific contexts, such as interaction logs (Brahim
(2022)), assessment grades, and online activity data
(Alhassan et al. (2020)), or based on teacher pedago-
gies (Trindade and Ferreira (2021)).
Thus far, the studies we cover demonstrate the
potential of data mining and machine learning tech-
niques to predict various student outcomes in educa-
tional settings. However, they also highlight chal-
lenges such as the need for extensive and diverse
datasets, the lack of validation of models on new
datasets, the lack of study on predicting SP at the pro-
gram level, and potential ethical concerns related to
using sensitive student data. Furthermore, the focus
on a single course at the course level did not lead the
educators to make a final decision on overall students’
performance at the end of the program or academic
year. Meanwhile, predicting students’ performance
at the program level is demanded. Based on this ob-
servation, our research examines the critical attributes
in LMS data relevant to predicting SP at the program
level. Focusing on the variables collected from Moo-
dle, our research investigates the relationship between
student engagement activities and SP, presented later
in section 3.2.1.
3 METHODOLOGICAL
APPROACH
3.1 Attributes Identification
The attributes identified in our study are derived from
an extensive literature review as previously presented
in section 2. We dedicated time to scrutinize vari-
ous papers that use different attributes to predict aca-
demic outcomes, success or dropout rates. We un-
covered six composite attributes commonly used in
predictive models: Demographic Details, Education
Information, Family Expenditure, Habits, College Fa-
cilities, and Teacher Pedagogy, as illustrated in Figure
1. These attributes guide teachers, providing a foun-
dational understanding to navigate the numerous at-
tributes available in an LMS and their relationships.
It is crucial to note that Figure 1 does not represent
an exhaustive list of all possible predictors of SP. In-
stead, it serves as a starting point to facilitate decision-
making for teachers in selecting pertinent attributes
from the multitude available in LMS when aiming to
predict SP. The intention is to offer researcher and
institutions the ability to decide when and which at-
Students’ Performance in Learning Management System: An Approach to Key Attributes Identification and Predictive Algorithm Design
287

Figure 1: The attributes for predicting SP.
tributes to use based on relevance, practicality, and
ease of interpretation within their specific context.
After a thorough and rigorous process of attribute
identification, we have gained valuable knowledge to
focus our effort on developing our predictive model.
Specifically, attributes presented later in section 3.2.1
have been prioritized from the broader set defined in
our model. This strategic selection aims to provide a
more practical and targeted approach for both teach-
ers and the research team at CADT and in France.
3.2 Predictive Approach
This research uses data mining techniques and predic-
tive algorithms within the Moodle environment. The
methodology adapted for this research is divided into
several steps, as shown in Figure 2.
Figure 2: The predictive approach.
3.2.1 Data Extraction and Preprocessing
The dataset used in our research comprises two pri-
mary sources: the Moodle LMS and Google Sheets.
The Moodle data has around 1000 students enrolled in
both short course training and a bachelor program, to-
talling around 5 million records. By conducting a pre-
liminary analysis, we kept only the bachelor’s degree
data because the data from the short course training
does not contain any assessment score or final score,
which is the critical target variable. Moreover, the stu-
dents from short course training had less interaction
with Moodle. Our study aimed to predict SP at the
program level. Among all registered students, many
were enrolled in short courses or workshops, which
did not provide a comprehensive view of their per-
formance within a program. Short courses typically
focus on specific topics and may only cover some as-
pects required to evaluate a student’s overall perfor-
mance in a program. To ensure that our analysis and
predictions were based on a thorough understanding
of SP in the learning environment, we focused on 160
students representing a diverse range of courses, in-
cluding Linear Algebra, Discrete Mathematics, Prob-
ability and Statistics, C Programming Language, Vi-
sual Art, Soft Skills, and Information Technology Es-
sentials. Because the short course training data could
not be used, we selected only 2 million records from
the bachelor program. Queries were used to count
the number of records in each attribute to obtain data
at the program level for both the first and second
DATA 2024 - 13th International Conference on Data Science, Technology and Applications
288

terms of the first-year program. Finally, we obtained
a dataset with 310 records from Moodle. Afterwards,
another set of 310 records was collected from Google
Sheets, representing this time the final scores of stu-
dents enrolled in both terms of their first year of the
bachelor’s degree, aligning with the data collected
from Moodle during the academic year 2022. The
dataset of two terms represents two program levels.
The attributes in the dataset provide information on
various aspects of a student’s engagement and perfor-
mance in all courses. The attributes are collected and
counted with queries are attendance, number of inter-
action log, total quiz submitted, total task submitted,
total assignment submitted, time spent on course, and
outcome score.
3.2.2 Selection Feature
Feature selection is an essential step in predicting SP
as it allows for the identification of the most rele-
vant attributes that contribute to the outcome. Several
feature selection methods are available, such as the
Pearson correlation coefficient, Spearman’s rank cor-
relation coefficient, mutual information, and recursive
feature elimination. Each technique offers advantages
and disadvantages depending on the dataset and the
problem being addressed.
Our study employs the Pearson correlation coef-
ficient as a feature selection method. By employ-
ing the Pearson correlation coefficient, the informa-
tion related to the education information category,
specifically engagement activities, as found in col-
lected data from CADT’s Moodle. Therefore, our
study will investigate the relationship between stu-
dent engagement activities and performance using the
attributes already mentioned in section 4.2.1. It is
worth reminding that these attributes are significant
predictors of student outcomes in the existing litera-
ture. However, there may be differences between the
attributes discussed in the literature and those in our
dataset. Nonetheless, our primary goal was to iden-
tify and analyze attributes that could be feasibly ob-
tained from Moodle while still providing valuable in-
sights into SP. In addition, by focusing on Moodle-
based data, we aimed to develop a model easily ap-
plied and adapted in similar LMS environments. Plus,
the selected Moodle-based attributes still capture es-
sential aspects of SP. Our findings can contribute to
the broader understanding of factors that impact aca-
demic success in online learning environments.
3.2.3 Classifier
To predict student outcomes and select the best clas-
sifier, several classification methods are used with our
dataset for comparison. As found in our studies in
related work, the most commonly used data mining
classifiers include Decision Trees, Random Forest,
Neural Networks, Naive Bayes, and Support Vector
Machine. In this research, we have made comparisons
of those five classifiers with our dataset.
3.2.4 Performance Evaluation Measures
To evaluate the performance of our classifier, we em-
ployed the confusion matrix, which is a widely used
as an evaluation measure in machine learning for
summarizing classifier performance. It provides in-
formation about the number of true positive, true neg-
ative, false positive, and false pessimistic predictions
made by the classifier. According to Aguiar et al.
(2014), this information is presented in a matrix for-
mat, where each row represents the actual class, and
each column represents the predicted class. The en-
tries in the matrix provide insight into the classifier’s
ability to predict each class and its tendency to mis-
classify instances. This evaluation measure is crucial
in unbalanced datasets, such as in this case, where the
number of failing students is much smaller than that
of successful students F
´
elix et al. (2018).
3.2.5 Student’s Performance Predictive
Algorithm
We have defined the algorithm formula for predicting
SP as a mathematical equation that uses a combina-
tion of various student attributes and their correspond-
ing coefficient values to calculate a predicted score.
The formula starts by taking the sum of all attribute
values and then normalizing each value by dividing
it by the range of possible values for that attribute.
Then, each normalized attribute value is multiplied by
its corresponding coefficient value, representing the
weight or importance of that attribute in the overall
prediction. Finally, the sum of all these weighted at-
tribute values is divided by the sum of all the coef-
ficients to arrive at a final performance score. This
score can then be used to classify students as likely to
succeed or likely to struggle in their studies.
sp =
N
∑
i=1
C
i
x
i
− min
i
max
i
− min
i
(1)
where





sp, x, C ∈ R
0 ≤ sp, C ≤ 1
∑
N
i=1
C
i
= 1
sp the student’s performance
x
i
the value of attribute i
C
i
the coefficient of attribute i
Students’ Performance in Learning Management System: An Approach to Key Attributes Identification and Predictive Algorithm Design
289

max
i
the maximum value of attribute i
min
i
the minimum value of attribute i
N the number of attributes
This approach for predicting SP is based on math-
ematical modeling and statistical analysis principles;
by using a weighted average calculation incorporat-
ing multiple variables, the algorithm can provide a
more accurate and reliable prediction of a student’s
likely performance. Moreover, the normalization of
attribute values ensures that all variables are treated
equally, regardless of their scales or units of measure-
ment.
4 EXPERIMENTAL RESULTS
The experimental results of our research on assisting
teachers in using predictive techniques to evaluate the
student’s performance at CADT are as follows:
Figure 3: The accuracy of each classifier.
The data from CADT are analyzed with five dis-
tinct classifiers (as previously stated in section 4.2.4)
to examine the dataset and assess each model’s accu-
racy. Based on the outcome shown in Figure 3, the
best two classifiers are the Decision Tree classifier,
with an accuracy of 87.22%, and the Random Forest
classifier achieved a higher accuracy of 89.44%.
While accuracy is a common and intuitive met-
ric for evaluating classifier performance, it has limita-
tions, especially in class imbalances or unequal mis-
classification costs. Performance evaluation metrics,
such as confusion matrices, provide a more detailed
and nuanced understanding of how well a classifier is
performing. Next, we used confusion matrix to eval-
uate the classifiers. Based on the results in Figure 4,
we demonstrated that both Random Forest and Deci-
sion Tree offer good accuracy. Therefore, for the pur-
pose of comparison, we selected the prediction classi-
fiers, which include Decision Tree and Random For-
est. However, the latter was chosen for predicting SP
due to its superior performance.
Figure 4: The comparison of classifiers accuracy of Ran-
dom Forest and Decision Tree.
4.1 Attribute Coefficients
At the end of the classifier performance evaluation,
we obtained the coefficients of feature values for
each attribute to illustrate its importance in predict-
ing SP when using decision trees and random for-
est classifiers, as depicted in Figure 5. The co-
efficients indicate the weight each attribute has in
the classifiers. For example, the coefficient val-
ues for the number of interaction log, and time
spent on course are higher for both classifiers, in-
dicating that these attributes are essential and im-
pactful in determining SP. On the other hand, the
coefficients for the attendance, total quizzes
submitted, total tasks submitted, and total
assignments submitted are relatively low because
they are already factored into the final grade. This
indicates that these attributes have little importance
in influencing SP prediction, even if some teachers
might utilize them outside of Moodle.
Figure 5: The coefficient of each attribute of Decision Tree
and Random Forest.
DATA 2024 - 13th International Conference on Data Science, Technology and Applications
290

4.2 Verification of Hypotheses Through
Regression Analysis Results
The results we have analyzed and presented in the
previous sections can serve as a valuable guide for
teachers, assisting them not only in selecting the key
attributes but also in understanding the relevance, de-
gree of importance and impact of each attribute in pre-
dicting SP. The last step in our study is to examine the
independent variables’ P-value and confidence inter-
val (CI) to verify our hypotheses using the regression
analysis results as shown in Table 1.
Table 1: Regression results.
Hypo. t-statistic P-value CI (95%)
H1 16.782 0.000 [2.269, 2.870]
H2 32.168 0.000 [5.133, 5.800]
H3 24.816 0.000 [3.899, 4.569]
Hypothesis 1. According to the regression results
in Table 1, the P-value for the independent variable
attendance with the dependent variable grade is
0.000. It indicates a statistically significant relation-
ship between attendance and SP. On top of that, the
confidence interval of [3.899, 4.569] suggests that the
effect of attendance on the total quiz submitted is pos-
itive and significant.
Hypothesis 2. Based on the regression results in
Table 1, the P-value for the independent variable
number of interaction log with the dependent
variable grade is 0.000, indicating a statistically sig-
nificant relationship between the two variables. Fur-
thermore, the confidence interval of [5.133, 5.800]
shows that the effect of number of interaction
log on grade is positive and significant.
Hypothesis 3. As illustrated in Table 1, the P-
value for the independent variable time spent on
course with the dependent variable grade is 0.000,
showing a statistically significant relationship be-
tween the two variables. However, the confidence
interval of [3.899, 4.569] indicates that the effect of
time spent on the course on SP is positive and signifi-
cant.
In conclusion, all three hypotheses are supported
by the regression results, with all independent vari-
ables showing a statistically significant impact on
their respective dependent variables. Our findings
suggest that the utilization of these key attributes and
the prediction algorithm can assist teachers at CADT
in assessing SP and identifying those at risk of not
performing well. Indeed, the empirical validation we
provided here can contribute to aspects beyond SP
prediction. For instance, by understanding student
performance and obtaining pertinent data for the anal-
ysis process, educators can adapt and personalize in-
terventions and support strategies for a more targeted
approach to student success. We hope that the in-
tegration of these findings into educational settings,
where data-driven decision-making is a powerful tool,
will transform our teaching practices, leading to better
academic performance and well-being for students.
5 CONCLUSION AND FUTURE
WORK
Our research aimed to assist teachers in identifying
key attributes and choosing prediction algorithms to
evaluate students’ performance. Following a rigorous
literature review, our research effort also included an
empirical study. Indeed, data from an authentic learn-
ing situation at CADT has been used in our first at-
tempt to gain a better understanding of predictive an-
alytics in education, which has become increasingly
important. On top of that, through our analysis, we
demonstrated the identified key attributes, as listed in
section 4.2.1, have a statistically significant impact on
student performance. We also determined that Ran-
dom Forest models are effective in predicting student
performance.
The major contribution of our research effort can
be summarized in two aspects. First, whereas pre-
vious studies have primarily emphasized the identi-
fication of factors influencing student outcomes, our
research goal is to explore and explain the impact of
the identified key attributes and their complex rela-
tionships. Second, the initial iteration of our predic-
tive algorithm is designed for both course and level
programs, while existing approaches from our liter-
ature review mainly focused on the course level for
predicting students’ outcomes. The first data anal-
ysis batches helped us not only determine what and
how important attributes are in predicting student per-
formance, but also understand in which areas we can
improve teaching practices and support students from
predicting their academic performance. However, it is
essential to point out that our research has some limi-
tations and further research and validation are needed.
Current and future work in this area include:
1. Expanding the Sample Size and Validation: we
are currently conducting a study that covers a big-
ger dataset from our partners in France. We also
expect to validate the results of our study by com-
paring them to other studies and testing the model
on new data to ensure its ability to generalize well
to unseen data.
2. Implementing the Model: our next challenge
Students’ Performance in Learning Management System: An Approach to Key Attributes Identification and Predictive Algorithm Design
291

will be a real-time application. For that, we are
studying the possibilities of integrating explain-
able AI methods like LIME or SHAP. As a matter
of fact, we are interested in helping our lecturer
colleagues interpret the model predictions.
3. Evaluation Metric: we acknowledge the value of
using Cohen’s Kappa for evaluating classifiers on
imbalanced datasets. Thus, we decided to incor-
porate Cohen’s Kappa as an additional evaluation
metric, alongside the confusion matrix. Indeed,
by using both evaluation approaches, we aim to
provide a more robust and comprehensive assess-
ment of the classifier’s performance in predicting
SP.
Overall, our research lays a foundation for ad-
vancing students’ performance prediction, benefiting
CADT and French partner universities and potentially
impacting the wider educational community. The
positive results suggest broader implications, influ-
encing global educational practices and fostering a
more data-informed and supportive learning environ-
ment.
REFERENCES
Aguiar, E., Chawla, N., Brockman, J. B., Ambrose, G. A.,
and Goodrich, V. (2014). Engagement vs perfor-
mance: using electronic portfolios to predict first
semester engineering student retention. Proceedings
of the Fourth International Conference on Learning
Analytics And Knowledge.
Albreiki, B., Zaki, N., and Alashwal, H. (2021). A sys-
tematic literature review of student’ performance pre-
diction using machine learning techniques. Education
Sciences.
Alhassan, A., Zafar, B. A., and Mueen, A. (2020). Predict
students’ academic performance based on their assess-
ment grades and online activity data. International
Journal of Advanced Computer Science and Applica-
tions, 11.
Arizmendi, C. J., Bernacki, M. L., Rakovi
´
c, M., Plumley,
R., Urban, C. J., Panter, A., Greene, J. A., and Gates,
K. M. (2022). Predicting student outcomes using dig-
ital logs of learning behaviors: Review, current stan-
dards, and suggestions for future work. Behavior Re-
search Methods, 55:3026 – 3054.
Brahim, G. B. (2022). Predicting student performance from
online engagement activities using novel statistical
features. Arabian Journal for Science and Engineer-
ing, 47:10225 – 10243.
Costaa, C., Alvelosa, H., and Teixeiraa, L. (2015). Cen-
teris 2012-conference on enterprise information sys-
tems the use of moodle e-learning platform : a study
in a portuguese university.
Felix, I., Ambrosio, A., Duilio, J., and Sim
˜
oes, E. (2019).
Predicting student outcome in moodle. In Proceed-
ings of the Conference: Academic Success in Higher
Education, Porto, Portugal, pages 14–15.
F
´
elix, I. M., Ambrosio, A. P., da Silva Neves Lima, P., and
Brancher, J. D. (2018). Data mining for student out-
come prediction on moodle: a systematic mapping.
Anais do XXIX Simp
´
osio Brasileiro de Inform
´
atica na
Educac¸
˜
ao (SBIE 2018).
Gaftandzhieva, S., Talukder, A., Gohain, N., Hussain, S.,
Theodorou, P., Salal, Y. K., and Doneva, R. (2022).
Exploring online activities to predict the final grade of
student. Mathematics.
Hirokawa, S. (2018). Key attribute for predicting student
academic performance. In International Conference
on Education Technology and Computer.
Namoun, A. and Alshanqiti, A. (2020). Predicting student
performance using data mining and learning analytics
techniques: A systematic literature review. Applied
Sciences.
Quinn, R. J. and Gray, G. (2019). Prediction of student
academic performance using moodle data from a fur-
ther education setting. Irish Journal of Technology
Enhanced Learning.
Ramesh, V., Parkavi, P., Ramar, K., Thai-Nghe, N., Busche,
A., Schmidt-Thieme, L., Carol, Bastin, P., Thiyagaraj,
Yosuva, S., and Arulkumar, V. (2013). Predicting stu-
dent performance: A statistical and data mining ap-
proach. International Journal of Computer Applica-
tions, 63:35–39.
Trindade, F. R. and Ferreira, D. J. (2021). Student perfor-
mance prediction based on a framework of teacher’s
features. International Journal for Innovation Educa-
tion and Research.
Venkatachalam, S. T., Doraisamy, S. C., Golzari, S.,
Norowi, N. B. M., Sulaiman, M. N. B., Liu, H., Mo-
toda, H., Setiono, R., Zharo, Z., and Maben, A. F.
(2011). Identifying key performance indicators and
predicting the result from student data. International
Journal of Computer Applications, 25:45–48.
Ya
˘
gcı, M. (2022). Educational data mining: prediction of
students’ academic performance using machine learn-
ing algorithms. Smart Learning Environments, 9.
DATA 2024 - 13th International Conference on Data Science, Technology and Applications
292