Content uploaded by Fidelia Orji
Author content
All content in this area was uploaded by Fidelia Orji on Nov 22, 2020
Content may be subject to copyright.
Using Machine Learning to Explore the Relation
Between Student Engagement and Student
Performance
Fidelia Orji, Julita Vassileva
Computer Science Department
University of Saskatchewan
Saskatoon, Canada
fidelia.orji@usask.ca, jiv@cs.usask.ca
Abstract—Engagement in learning activities is an important
factor that affects student performance in education. According
to research, student engagement involves the degree of passion,
interest and attention that they exhibit in their educational
environment. In the traditional learning system, educators
encourage students to engage in their learning activities through
various teaching strategies such as making them pay attention,
take notes, ask questions and participate actively in the learning
processes. Sometimes, educators call on a specific student to
answer a question as a means of encouraging the student to
participate in learning processes. Nowadays, engagement
strategies for learning are changing, especially with the use of
technology-enhanced learning systems (TELS) in education. As
a result, improving the engagement level of students in online
learning environments remains an open research question that
needs to be explored. This research is part of a preliminary
study on discovering ways of increasing student engagement in
an online learning system through data-driven interventions.
Student engagement in this research is determined using
objective data (activity logs of a specific undergraduate course
in a TELS). Activity log is unbiased data and a reflection of
students' actual learning behaviours (uncontrolled). In this
study, we mined the log of students’ learning activities from a
TELS used for an undergraduate course to explore differences
between students’ learning behaviours as they relate to their
engagement level and academic performance (measured in
terms of final grade points in a course). We employed supervised
(Random Forest) and unsupervised (Clustering) machine
learning approaches in exploring the relations. The approaches
identified an interesting pattern on student engagement and
show that engagement and assessment scores are good
predictors of student academic performance. Assessment scores
are measured with results of quizzes and assignments
performed by the students in the TELS, while academic
performance is measured with the final grade of the student in
the course. The implications of our findings are discussed.
Keywords—student engagement, student performance, machine
learning, supervised and unsupervised machine learning,
clustering, random forest, educational data mining, learning
pattern, online learning, academic performance, technology-
enhanced learning systems.
I. INTRODUCTION
Increasing use of online learning systems nowadays for
both eLearning courses and blended learning (a combination
of face-to-face teaching with web-based TELS) has resulted
in the generation of a huge volume of learning data. Research
in learning analytics is harnessing the data to understand the
real learning behaviour of students and determine factors that
improve learning success. Increasing attention is paid to
student engagement in learning as one of the factors affecting
student performance. Various research studies revealed the
importance of student engagement in both face-to-face
teaching [1] and online learning systems [2]. The resulting
theoretical models on student engagement especially in higher
education have formed the basis for discussion about the
relation between learning engagement and other learning
factors such as student performance.
Previous studies investigating student engagement and
performance in learning usually take the form of surveys in
measuring engagement. In survey-based studies, students
typically answer questions designed using existing models for
student engagement. This approach uses self-reports, which
are often biased and subjective, hence the results may not be
realistic. However, student engagement for learning could be
determined using objective data such as logged students’
activities in learning-related tasks. According to research,
mining data of students’ learning logs could reveal their real
learning behaviour which may help in identifying patterns of
learning that are successful [3]. Thus, analyzing learning
systems data will help in determining students' actual learning
behaviours and in reporting their learning progress which will
assist educators in their decision-making process. The analysis
could also support assessing the relation between learning
variables (for example student engagement) and student
performance.
Previous research pointed to the need to use data from the
students’ learning behaviours and characteristics in predicting
their performance [4], especially in TELS. In this study, we
mined students’ learning logs to gain insights about the
relationship between their engagement level and their
academic performance. We used a dataset collected from a
blended learning system used in a large first-year
undergraduate class at our University. The dataset provided
information on students’ activities and interactions with the
TELS. We performed an exploratory analysis and applied
unsupervised and supervised machine learning methods to
students’ activities to determine how they affect their
academic performance.
This research seeks to find the relation between
engagement variables, segment the variables and assessment
scores using cluster analysis, explore the characteristics of the
different segments, and predict student performance using the
engagement variables and assessment scores.
Specifically, the goal of this study is to answer the
following specific research questions:
RQ1: How do the student engagement variables relate to
each other? Are there identifiable groups of students with
certain patterns of engagement variables values that perform
better?
RQ2: What is the relationship between student engagement
variables and their assessment scores?
RQ3: What is the relationship between student engagement
variables, assessment scores and actual academic
performance?
This research adds to the existing studies on the use of
learning analytics in understanding students learning
progress and in supporting educational institutions in making
appropriate decisions that will improve students’ learning. It
also adds to existing research on student engagement in
higher education with new insight obtained from learning
from log data.
II. BACKGROUND
A. Student Engagement for Learning
According to student involvement theory for higher
education, the learning and personal growth of students in an
educational program increase as the quality and quantity of
students’ involvement increase [5]. The theory postulates that
involvement could be a quantitative measure such as time
spent on learning activities or qualitative such as measures of
learning goals. The measures could be general or specific (that
is involving entire student experience in learning or just
experience in a specific course). Based on the theory, highly-
involved students devote considerable energy and time in
studying and participating in academic activities. Moreover,
research suggested that universities that highly engage their
students with a variety of relevant learning activities that help
to improve the learning outcome of their students may be
considered to have a higher learning quality than a university
with less engaging activities for students’ learning [6]. This is
because the more students study, practice, perform
assessments and get feedback, the deeper their understanding
of what they are learning.
Research has studied student engagement in education in
various forms. For instance, Pace [7] one of the earliest
researchers on student engagement developed the College
Student Experience Questionnaire (CSEQ) tool. Pace reported
that students that devoted more time and energy to learning
tasks gained a lot from their studies in terms of college
experience and application of concepts learned to concrete
situations. There is growing importance in understanding the
effect of student engagement on their learning experience and
to institutions of learning. Various communities such as
Community College Survey of Student Engagement (CCSSE)
and National Survey of Student Engagement (NSSE) have
been developed for assessing the quality of effort and
participation of students in useful learning activities. In line
with the relevancy of student engagement, research
highlighted the role of institutions in improving engagement
as it affects institutions' and students’ performance [6].
As online education continues to penetrate both blended
and distance learning systems, the need for improving
students’ learning experience and performance in online
systems become a vital issue to explore. Various researchers
have studied student learning experience and engagement
using different survey-based approaches. For example,
Delfino [1] in a survey-based study investigated factors
affecting student engagement and its association with
academic performance using statistical methods. However,
few studies exist on exploring student engagement using their
actual learning behaviour in an uncontrolled learning system
(a system where students can log in and study to meet their
set learning goals at will). Thus, this research studies student
engagement using their actual learning activities logs and
machine learning approach.
B. Machine Learning Algorithms and Educational Datasets
Intelligent educational systems learn from student
activities interaction data and adapt/improve/personalize their
strategies and content. They use data mining techniques from
supervised and unsupervised learning algorithms. The
educational data mining (EDM) area has a more general focus;
it explores datasets generated from students’ learning
activities using different machine learning and data mining
algorithms to understand students’ learning processes and
their learning environment [8]. With the help of the
algorithms, researchers have been able to find answers to
specific problems concerning students’ learning experience
and effectiveness. For instance, in identifying students that are
likely to fail a particular course using the students’ previous
performance data and decision tree algorithm, a predictive
model was built using engineering students’ data [9]. The
model was used to detect in advance the students that are
likely to fail a course so that adequate assistance for
improvement of their learning could be provided for them. On
the other hand, in predicting students that will likely proceed
to pursue a postgraduate degree, a study collected data from
senior undergraduate students with the use of questionnaire
and applied decision tree algorithm in Weka, the result
showed a classification accuracy of 88% [10].
Studies have made efforts to analyze the learning
interaction of students in various systems to obtain insights
concerning different students’ learning approaches and to
answer some research questions based on specific goals. Some
of the studies try to model students based on their learning
behaviour. For example, Amershi et al. [11] built a framework
with both supervised and unsupervised classification
algorithms for identifying useful learning interaction of
students. The framework was applied to two different
environments of learning using logged and eye-tracking data.
The authors suggested that their framework could be used for
automatic classification of learning behaviours of new
students on online learning systems. Many other works
demonstrate how artificial intelligence techniques and
statistical tools can be applied in evaluating and adapting e-
learning systems to students [12]. For example, the usage
patterns of learners on the e-learning system can be classified
according to usage level for the purpose of adapting the
content and structure of the e-learning system and also for
detecting learners that are not regular.
Most higher institutions of learning use course
management and e-learning systems for posting and providing
access to course materials for students. According to research,
these systems do not offer educators the opportunity to
evaluate learning processes and course effectiveness based on
activities performed by students [13]. Thus, several studies
providing insight from educational data through the use of
clustering algorithms have been performed. Parack et al. [14]
in a study on profiling and grouping students based on their
academic records, applied apriori algorithm to students’
academic records to extract association rule for profiling and
the k-means algorithm was used in grouping the students
based on their learning pattern. They reported that their
implemented algorithms could provide an efficient way of
profiling students. Similarly, research on improving
accessibility of learning objects through a personalized
learning setting proposed a combination of k-means algorithm
and self-organizing map for clustering and ranking learning
objects [15]. Furthermore, the Expectation-Maximization
(EM) clustering algorithm is frequently used for the clustering
of data in machine learning. The algorithm has been applied
to educational data for various purposes. For instance,
research has shown that the application of EM to course
evaluation data discovered useful student profiles [16].
Bogarin et al. [17] proposed a model that first applied the EM
algorithm to group students on basis of their performance and
based on the result of the clustering, students’ behaviour for
each cluster was discovered. A review of various applications
of clustering to educational datasets for different purposes is
provided in [18].
III. METHODOLOGY
We performed this study to identify different groups of
students with important characteristics related to their
learning and to predict student performance using objective
data. To answer the research questions, we performed some
exploratory analysis and report our results over the same set
of features.
A. Data Collection and Processing
The data used for this research was collected in a blended
learning course (Biology) taken by undergraduates in a
Canadian university. The students involved in the course used
a TELS called MindTap system [19]. The system logged data
on students’ actions, activities, and assessments.
To obtain some relevant features that might assist us in
determining engagement level and assessment scores of
students, we cleaned and prepared the dataset using Python.
We removed some features that might not be relevant to our
analysis. Furthermore, students’ records without logged
activities and actions (null data) were deleted. After the data
preprocessing, we were left with data (records of students’
activities) from four hundred and ninety (490) students. Some
of the features selected from the dataset for the analysis
include the following:
Total time spent in MindTap (TimeOnTask) – This feature
shows the total amount of time that each student has spent in
MindTap on various activities such as Homework,
Assignments, Quizzes, and Readings. The time was logged in
hours, minutes and seconds. We converted the total time to
minutes as there was nothing logged on seconds.
Number of logins (NumberofLogins) – This displays the total
number of logins in MindTap for each student.
Percentage of Activities Accessed (ActivitiesAccessed) –
This indicates the percentage of activities accessed by each
student out of the total number of activities assigned.
Overall Score in percentage (AveAssessmentScore) – It
indicates the average performance score for each student
based on the score of all relevant assessments performed on
the MindTap system.
Furthermore, we explored the dataset to get information
on the distribution of values within each of the selected
features. Table 1 and Figure 1 gives information on the
description and distribution of the features on our dataset.
Table 1 shows the total number of student records on the
dataset as 490 and other statistics about each feature. For
example, the mean of NumberofLogions is 32.5, the
minimum is 1.0, the maximum is 186,0 and the 25th, 50th, 75th
percentiles are 21, 29, and 40 respectively. Figure 1 helps us
to determine whether the distribution of values within the
features are different.
TABLE 1. SUMMARY STATISTICS OF OUR DATASET
Fig. 1. Distribution of each feature in our dataset
As can be seen from Figure 1, two of the features
NumberofLogins and TimeOnTask are skewed right (their
tails extend towards the right). The figure shows that the two
features contain outliers. The feature ActivitiesAccessed is
roughly symmetric. The assessment performance
AveAssessmentScore is left-skewed. The figures show that
the distribution of values within each feature is different.
Based on the result of our dataset exploration, the outliers
were deleted and a total of four hundred and eighty-eight
(488) students’ records were used for the analysis. Also,
approaches that will optimize the distribution of the dataset
features were chosen for the analysis.
B. Data Analysis
To determine the degree of association between the
selected engagement variables (learning activities features),
we performed a correlation analysis to measure the
relationship between the engagement variables using the
Spearman correlation coefficient in Python. The result is
shown in Table 2.
In determining different groups of students based on their
engagement variables, we applied clustering, an
unsupervised machine learning method suitable for
partitioning data meaningfully to discover hidden patterns in
it. The clustering used the Expectation-Maximization (EM)
[20] algorithm as implemented in Weka. The algorithm uses
a random initialization and iterative process which alternates
the expectation, E and maximization, M steps continuously
until the algorithm convergence [21]. It tries to optimize the
parameters of the model to best explain the dataset through
the maximization of the likelihood of the data in the final
clusters. Research has shown that the EM algorithm is useful
when using a real-world dataset that involves clustering small
scenes (features) where k-means cannot perform well [22].
Several studies that proposed students’ modelling and
profiling via a data-driven approach have used the EM
algorithm in achieving various goals concerning students
learning [16], [17]. The algorithm instead of trying to
maximize the difference in mean of data instances maximizes
the likelihood of a given data in the final cluster using
computation of the likelihood of cluster membership based
on probability distribution. The algorithm has the advantage
of approximating the observed distributions of features
according to mixtures of different distributions in the clusters
and it automatically determines the appropriate number of
clusters. This process of hyperparameter tuning of the
algorithm helps in determining the optimal number of clusters
for a given clustering problem. The result of the clustering is
shown in Table 3.
Predicting Academic Performance of Students
Having gained insight on the relationship between
engagement variables and assessment performance through
unsupervised machine learning approach – clustering, we
decided to investigate the degree of association between
engagement variables and academic performance (final grade
in a course) of students. We employed a supervised machine
learning algorithm called random forest in investigating the
impact of engagement and assessment scores on academic
performance. The random forest algorithm is a good option
when features in a dataset are not well scaled. It performs
classification and regression tasks. For this study, we applied
the random forest algorithm for a regression task. The
algorithm is very stable and it has reduced bias because it
combines multiple decision trees through an ensemble
learning method and builds trees using random data points
from the training set. The ensemble learning uses bagging
technique and this allows individual decision trees (subsets) to
run in parallel without interacting with each other. The
algorithm uses the average outcome of each tree in predicting
its final outcome and this helps to improve its prediction
performance and prevents overfitting through random
sampling of data subsets.
Using Scikit-Learn implementation of the random
forest algorithm in Python, we constructed a model that can
predict students’ academic performance in a university
course based on their engagement variables and assessment
scores. We applied percentage split technique to our dataset,
80% for training and 20% as a test set. To find the number of
trees parameter value that can best predict academic
performance, we performed hyperparameter tuning. The
number of trees parameter was optimized based on root mean
squared error (RMSE). The parameter values tested were 10,
20, 30, 40, 50, 60, 100, 200, 500, and 1000. We obtained
optimal parameters setting when the number of trees
parameter was set to 40, the random state to 42 and the other
parameters used their default settings. The model was then
evaluated using the test set to determine how it will perform
on a new dataset. The result of the prediction is presented in
the next section.
IV. RESULTS AND DISCUSSION
A. The Relation between Engagement Variables
The results of the Spearman correlation in Table 2 show
that the three engagement variables used in this research:
ActivitiesAccessed, TimeOnTask, and NumberofLogins
have a positive correlation among them. The positive
correlation indicates that the variables will likely perform
well as engagement measures. This answers our research
question on the relation between the engagement variables.
TABLE 2. SPEARMAN CORRELATION RESULT FOR
ENGAGEMENT VARIABLES
ActivitiesAcc
essed
TimeOnTask NumberofLogins
ActivitiesAccessed 1.0000 0.5246 0.4533
TimeOnTask 0.5246 1.0000 0.6002
NumberofLogins
0.4533 0.6002 1.0000
B. The Relationship between Engagement Variables and
Assessment Scores
The application of clustering to our dataset identified
interesting students’ categories as clusters. Each of the
clusters significantly differs in their characteristics as shown
in Table 3. The three clusters created were labelled as C0 for
the first cluster, C1 and C2 for the second and third clusters
respectively. Students grouped in C0 (148 students) were
highly engaged as shown by the measures of engagement
variables (ActivitiesAccessed, TimeOnTask, and
NumberofLogins) and they had an excellent performance
(89.561) as indicated in their assessment measure. The
students in this group are assumed to have adopted a
dedicated approach to learning which consequently affected
their assessment performance. For C1, the students in this
group (116 students) were not actively engaged as indicated
in their engagement measures. They did not show much
commitment to their learning activities and it affected their
assessment performance (56.104). The students grouped in
cluster C2 (224 students) were more committed to their
learning activities and they performed better than those in C1.
In answering one of our research questions, we can say that
the students in cluster C0 performed better than those in the
other two groups. This means that the higher the engagement
for learning activities, the better the assessment scores. This
result is consistent with other studies in literature that revealed
that student performance relates to their level of engagement
[23]. Moreover, the result shows that the C0 group that was
highly engaged performed better in assessments than the
others who were not deeply engaged.
TABLE 3. CLUSTERING RESULTS OF THE EM
ALGORITHM
Clusters
Features C0 (Mean) C1 (Mean) C2 (Mean)
ActivitiesAccessed 9.891 5.644 7.000
TimeOnTask 1886.597 654.602 1128.790
NumberofLogins 45.114 18.838 30.035
AveAssessmentScore 89.561 53.413 85.948
The number of students in each cluster is as follows: C0 has
148 students (30%), C1 contains 116 (24%), and C2 contains
224 (46%).
C. The Relationship between Engagement Variables,
Assessment Scores and Actual Academic Performance
The result of our random forest model shows that there is
some relation between our selected features (engagement
variables and assessment scores) and the students’ actual
academic performance. The evaluation result of the model on
the test set shows an accuracy of 84.10% and root mean square
error (RMSE) of 12.35. Accuracy was calculated using the
mean absolute percentage error.
To determine the usefulness of each feature in improving
the model, we checked the relative importance of the features
using Scikit-Learn. The result shows features importance as
follows: AveAssessmentScore contributes 60%,
TimeOnTask contributes 20%, NumberofLogins contributes
13%, and ActivitiesAccessed contributes 7%. The
assessment scores (AveAssessmentScore) is the highest
contributing factor, followed by time on task (TimeOnTask)
and the percentage of activities accessed
(ActivitiesAccessed) is the least.
D. The Implications of our Results
Mining learning logs of students’ activities could provide
useful information for profiling and grouping them based on
their learning patterns. Research has shown that students have
different learning characteristics that affect their ability to
learn. Thus, grouping students with similar engagement levels
will provide an interesting way of tailoring learning
interventions to students based on their engagement needs.
Appropriate interventions optimizing learning of the different
levels could be provided. Such intervention might involve the
use of both internal and external motivators such as
visualizations, incentive mechanisms or persuasive
technology in encouraging students to actively participate in
their learning activities. These approaches could be applied in
TELS using learning data as they have been shown to improve
participation. For example, research has shown that presenting
different levels of contribution of users in an online
community using visualization has a significant effect on
improving participation [24]. Consequently, automating the
grouping on technology-enhanced learning systems (TELS)
using clustering model as shown in this research, and
reporting the data using visualizations that educators
understand, will help in providing useful information on the
progress of learners. The information will assist educators in
determining if the students are deeply involved in their
learning activities. If it is found that the students are not
committed to their learning as they should, the influencing
factors (such as design, structure and pedagogical elements of
the TELS) could be investigated and this will assist
institutions in taking proper decisions on improving students
learning experience and performance.
Our prediction model in this research has shown that
engagement levels and assessment scores of students in TELS
are good predictors of their academic performance. With the
use of this model in TELS, individual students can be
presented information on how their study practices and
assessments affect their performance and this will increase
their awareness of what their final grade will be if they do not
improve in their study practices. Moreover, the model will
assist educators in identifying on time students that are likely
to fail/drop (at-risk students) a course. Hence, appropriate
measures for helping at-risk students could be initiated
automatically without much resources from educators which
will help to save resources for other purposes. According to
research, improving student engagement could help
educational institutions in addressing problems of high
dropout rate, low performance and boredom among students
[25].
V. CONCLUSION
Student engagement as a vital construct in understanding
student learning behaviour could be used in evaluating
technology-enhanced learning systems on their ability to
properly impact students’ learning especially now that higher
education institutions incorporate TELS as part of the required
learning medium for students. The data from these systems
provide information on how the students engage with them to
achieve their learning goals. Analysis of the data provides
educators with reliable information on students’ learning
progress which will help them in identifying students learning
needs and in making decisions on how to improve the learning
experience of students.
This paper presented preliminary work on students' group
modeling based on their learning interaction to gain an
understanding of how their engagement indicators on TELS
affect their academic performance. It applied machine
learning methods to educational data obtained in a blended
learning environment to achieve its goal. The work
highlighted the relationships between engagement level and
student academic performance and how machine learning
algorithms could help educators in monitoring and responding
to students’ learning progress issues automatically, thereby
allowing them to spend their time on other pedagogical issues.
Higher education institutions could apply the group
modeling approach in this research in detecting how effective
a TELS is at inspiring students for learning and also in
offering automatic adaptive interventions based on this group
modeling which might be difficult to accomplish for
individual students (using the predictive model). The
adaptivity of the systems will be in response to observed
pattern of learning needs.
REFERENCES
[1] A. P. Delfino, “Student engagement and academic performance of
students of Partido State University,” Asian J. Univ. Educ., vol. 15,
no. 1, pp. 22–41, 2019.
[2] H. J. Kim, A. J. Hong, and H. D. Song, “The roles of academic
engagement and digital readiness in students’ achievements in
university e-learning environments,” Int. J. Educ. Technol. High.
Educ., vol. 16, no. 1, p. 21, Dec. 2019.
[3] G. McCalla, “The Ecological Approach to the Design of E-
Learning Environments: Purpose-based Capture and Use of
Information About Learners,” J. Interact. Media Educ., vol. 2004,
no. 1, p. 3, May 2004.
[4] M. Vahdat, A. Ghio, L. Oneto, D. Anguita, M. Funk, and M.
Rauterberg, “Advances in Learning Analytics and Educational
Data Mining,” in 23rd European Symposium on Artificial Neural
Networks, Computational Intelligence and Machine Learning,
ESANN 2015 - Proceedings, 2015, pp. 297–306.
[5] A. W. Astin, “Student involvement: A developmental theory for
higher education,” J. Coll. Stud. Dev., vol. 40(5), pp. 518–529,
1999.
[6] G. D. Kuh, “The national survey of student engagement:
Conceptual and empirical foundations,” New Dir. Institutional
Res., vol. 2009, no. 141, pp. 5–20, 2009.
[7] C. R. Pace, “Measuring the quality of college student experiences:
An account of the development and use of the college student
experiences questionnaire,” High. Educ. Res. Inst., pp. 1–136,
1984.
[8] C. Romero and S. Ventura, “Educational data mining: A review of
the state of the art,” IEEE Transactions on Systems, Man and
Cybernetics Part C: Applications and Reviews, vol. 40, no. 6. pp.
601–618, Nov-2010.
[9] B. R. S. Kabra R R, “Performance Prediction of Engineering
Students using Decision Trees,” Int. J. Comput. Appl. (0975 -
8887) Vol. 36- No.11, December 2011, vol. 36, no. 11, 2011.
[10] V. P. Breşfelean, “Analysis and predictions on students’ behavior
using decision trees in weka environment,” in Proceedings of the
International Conference on Information Technology Interfaces,
ITI, 2007, pp. 51–56.
[11] S. Amershi and C. C. Conati, “Combining Unsupervised and
Supervised Classification to Build User Models for Exploratory
Learning Environments,” JEDM-Journal Educ. Data Min., vol. 1,
no. 1, pp. 1–54, Nov. 2009.
[12] R. Agrawal and R. Srikant, “Fast Algorithms for Mining
Association Rules in Large Databases,” in Proceedings of the 20th
International Conference on Very Large Data Base, 1994.
[13] M. E. Zorrilla, E. Menasalvas, D. Marín, E. Mora, and J. Segovia,
“Web usage mining project for improving Web-based learning
sites,” in Lecture Notes in Computer Science (including subseries
Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics), 2005, vol. 3643 LNCS, pp. 205–210.
[14] S. Parack, Z. Zahid, and F. Merchant, “Application of data mining
in educational databases for predicting academic trends and
patterns,” in Proceedings - 2012 IEEE International Conference
on Technology Enhanced Education, ICTEE 2012, 2012.
[15] A. S. Sabitha and D. Mehrotra, “User centric retrieval of learning
objects in LMS,” in Proceedings of the 2012 3rd International
Conference on Computer and Communication Technology,
ICCCT 2012, 2012, pp. 14–19.
[16] E. Trandafili, A. Allkoçi, E. Kajo, and A. Xhuvani, “Discovery and
evaluation of student’s profiles with machine learning,” in ACM
International Conference Proceeding Series, 2012, pp. 174–179.
[17] A. Bogarín, C. Romero, R. Cerezo, and M. Sánchez-Santillán,
“Clustering for improving Educational process mining,” in ACM
International Conference Proceeding Series, 2014, pp. 11–15.
[18] A. Dutt, “Clustering Algorithms Applied in Educational Data
Mining,” Int. J. Inf. Electron. Eng., 2015.
[19] “MindTap - The leading digital learning tool – Cengage.” [Online].
Available: https://www.cengage.com/mindtap/. [Accessed: 07-
Aug-2020].
[20] A. P. Dempster, N. M. Laird, and D. B. Rubin, “ Maximum
Likelihood from Incomplete Data Via the EM Algorithm ,” J. R.
Stat. Soc. Ser. B, vol. 39, no. 1, pp. 1–22, Sep. 1977.
[21] G. Celeux and G. Govaert, “A classification EM algorithm for
clustering and two stochastic versions,” Comput. Stat. Data Anal.,
vol. 14, no. 3, pp. 315–332, Oct. 1992.
[22] N. Sharma, A. Bajpai, and R. Litoriya, “Comparison the various
clustering algorithms of weka tools,” Int. J. Emerg. Technol. Adv.
Eng., vol. 2, no. 5, pp. 73–80, 2012.
[23] H. Lei, Y. Cui, and W. Zhou, “Relationships between student
engagement and academic achievement: A meta-analysis,” Soc.
Behav. Pers., vol. 46, no. 3, pp. 517–528, 2018.
[24] J. Vassileva and L. Sun, “Evolving a Social Visualization Design
Aimed At Increasing Participation in a Class-Based Online
Community,” Int. J. Coop. Inf. Syst., vol. 17, no. 04, pp. 443–466,
Dec. 2008.
[25] J. A. Fredricks, P. C. Blumenfeld, and A. H. Paris, “School
engagement: Potential of the concept, state of the evidence,”
Review of Educational Research, vol. 74, no. 1. pp. 59–109, 2004.
