Volume 8, Issue 11, November 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Optimizing E-Learning Platforms using

Machine Learning Algorithms
Peter N.Mulei1, Salesio Kiura2
1,2
School of Computing and Information Technologies,
Technical University of Kenya, P.O Box 52428-00200–Nairobi-KENYA

Abstract:- The proliferation of e-learning platforms and has a different learning style. Numerous approaches and
blended learning environments has spurred a great deal structures exist for customized education. Owing to the
of study on how to improve educational processes. The proliferation of networks and technical advancements, e-
problem with the e-learning platforms, give the content learning systems that support instructors and students in
as whole without considering the level of cognitivity of achieving individualized learning are desperately needed.
learners.One key factor is being able to forecast student One of the most well-liked individualized learning strategies
performance with accuracy. Early in the learning is learning styles, which take learners' learning preferences
process, it is useful to detect low-performing pupils into account [14]
based on a high forecast accuracy of their performance.
But in order to accomplish these goals, a lot of student II. RELATED WORK
data needs to be examined and forecast using a variety of
machine-learning models. Machine learning algorithms A. Introduction.
have shown to be a useful tool for focusing performances Predicting students' academic success is turning into a
at different learning levels when used to forecast crucial function in the computer-supported intelligent
learners' actions based on their performance and education system. But in order to forecast students'
background. For the purpose of enhancing learning academic achievement, traditional machine learning-based
techniques can only take use of the sparse discriminative
outcomes, early student performance prediction is
characteristics of students' behaviour in unbalanced
helpful. utilizing clever and flexible components to
provide students with a personalized learning academic datasets[5]But the majority of studies on AI
environment.Differentiating prediction levels by prediction models that have been conducted have mostly
different machine-learning models may be the result of focused on the development and optimization of AI models,
variations in socioeconomic conditions. Specialized scope i.e., using multiple algorithms to develop models with
classifiers are then merged into an ensemble to robustly higher prediction accuracy; this is in contrast to applying AI
forecast student achievement on learning objectives models to provide real-time and continuous feedback and
independently of the student's specific learning settings. improve the quality of student learning[8].
Personalized Learning Environmentsimprove the A number of research groups have emerged as a result
educational process by offering specialized services that of the greater usage of educational data, including learning
are based on the preferences of the learners. analytics, which aims to predict learners' behaviour and find
Keywords: Machine Learning Algorithms, E-learning, ways to inspire them while they learn. Gathering, compiling,
Ensemble, Personalized, and Adaptive. reviewing, and analysing student login records is one aspect
of learning analytics. This process allows for the inference
I. INTRODUCTION of the learning environment and optimizes the performance
of both lecturers and students(Viberg et al., 2018; Capuano
Machine learning is becoming widely used as a result & Toti 2019). With the use of technology-enhanced learning
of the quick advancement of technology. Learners' logs are platforms and an abundance of available educational data, it
recorded in electronic learning environments, and a is possible to mine student learning behaviour, correct
multitude of learning attributes are noted. The e-learning problems, optimize the learning environment, and facilitate
logs are analyzed using machine learning techniques to data-driven decision-making. Virtual learning environments
forecast how well the students will learn[6] Through the are a valuable addition to the learning analytics paradigm as
process of forecasting students' progress, instructors can step they efficiently furnish statistics for examining, reporting,
in early to prevent students from failing and increase their and reflecting on students' learning experiences and how
chances of achieving success. The three primary algorithms those experiences impact their individual performances.
in machine learning are classification, clustering, and Instructors can have real-time access to student information
regression[4] by using machine-learning algorithms, which enables them
to step in early in the course. Predictive models are
E-learning can be facilitated by the Learning frequently constructed using machine learning algorithms
Management System (LMS). Algorithms for machine and student data. With these machine approaches,
learning have been utilized to help provide educational categorical and numerical predictor variables can be
materials to students. Additionally, it enables instructors to handled.
evaluate and comprehend how pupils behave and perform as
they progress through the learning process. In e-learning,
personalization highlights how every learner is different and

IJISRT23NOV751 www.ijisrt.com 2670

Volume 8, Issue 11, November 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
B. Machine Learning Algorithms in E-learning Platforms of learning (e.g. facial recognition on videos or
Based on the kind of learning signal or feedback that a photographs)
learning system has access to, machine learning systems can
be broadly divided into four categories: supervised learning, H. Current machine Algorithms used in e-learning
unsupervised learning, semi-supervised learning, and platforms.
reinforcement learning. In order to provide meaningful
information, machine learning may extract and analyze data,  Decision Tree
search for patterns and correlations from huge data sets, and Utilizing a predictive machine-learning model, it
more. Several methods employed in the development of e- determines the value of a new sample by utilizing different
learning platforms include Neural networks, which consist attribute values from the accessible data. A decision tree's
of an interconnected network of nodes that collaborate to internal nodes represent the various qualities, its branches
provide solutions, Utilizing Support Vector Machine (SVM) indicate potential values for these attributes in the observed
to classify data in order to produce the optimal solution samples, and its terminal nodes indicate the dependent
space In order to identify a method for identifying the most variable's ultimate value (classification)[3]For learners'
appropriate route that leads to problem solving, decision datasets to be subjected to predictive analysis throughout the
trees generate images of decisions [13] learning process, data mining is required. Decisions
regarding the learners' future steps will be made using the
C. Algorithms in E-learning Systems. information or knowledge that data mining techniques can
A learning management system e-learner's likelihood of produce, which is the prediction of learner performance and
finishing the registered course must be predicted by a attainment[3]
machine. The machine uses its prior knowledge and
experiences to complete the assignment. Learner behaviour  Naïve Bayes algorithm.
data collected during the course, including interactions, It is a supervised learning approach for classification
responses to activities, and outcomes. The creation of issues that is based on the Bayes theorem. Its primary
algorithms that can take in input data, analyze it statistically, application is in text categorization, where a high-
and forecast an output while updating those outputs as new dimensional training dataset is used. One of the most
data becomes available is the fundamental idea of machine straightforward and efficient classification methods is the
learning[2]. Naïve Bayes classifier, which aids in the rapid development
of machine learning models with rapid prediction
D. Supervised Learning capabilities. It may be used to generate data from any
Accepting labeled examples such as an input where the institution utilizing data mining techniques for real-time
desired outcome is provided—is the first step in the creation predictions and the prediction of student performance and
of supervised learning. It offers a dataset with labels in achievement, which will help learners make decisions[3]
addition to features. This method is frequently applied in
situations where past data indicates the likelihood of future  Logistic Regression
events. Supervised learning falls into two categories: This categorization may be used to forecast distinct
regression and classification. Regression uses a continuous results. Its advantages include efficiency and a lower
label, whereas classification uses a discrete label. tendency to overfit. It is one of the supervised learning
algorithms that is helpful in statistical techniques and to
E. Unsupervised learning. predict the outcome of dependent variables. It is also the
Unsupervised learning makes use of data without prior simplest algorithm that doesn't require a lot of processing
examples with the aim of using data exploration to identify power, straightforward to build, and quite quick. The
similarities between the items. Actually, it's a method of logistic regression algorithm has a very low overfitting rate
identifying labels directly from the data. The prediction and is highly efficient. This algorithm is accurate and
model is created by the unsupervised learning algorithm basic[12] and it is simple to update to take into account fresh
during the training phase, and it attempts to fit its parameters data. The fact that it cannot solve nonlinear issues is one of
to the best summarized regularities it can find in the data. its drawbacks. This algorithm may cause the training dataset
to become overfit when it uses a high dimensional
F. Reinforcement Learning. dataset[12]
The algorithm of reinforcement learning finds out via
trial and error which action yields the highest reward.  Random Forest (RF)
Finding the best policy is the objective. In this kind of It is an ensemble approach with numerous trees that is
learning, the learner, the surroundings, and the activities are based on decision trees. Tree votes are the classifications
the three key elements. that every tree in the forest offers. The vote tree with the
highest vote is selected as the final forecast by the forest
G. Semi-supervised learning fusion. Because so many decision trees are utilized in the
Both labeled and unlabeled data are used in semi- process, random forest is an effective and extremely
supervised learning, and it is ideal for the task to already accurate strategy(Lassaad K. et al., 2022)Out-of-bag data is
have a "prediction problem." The model then gains the the remaining data that does not have a decision tree and is
ability to arrange data and predict as well. Classification, used to evaluate the decision tree's performance. Using
regression, and prediction techniques are applied to this kind feature ranking, which takes into account a feature's
significance across the entire dataset, it adopts the feature

IJISRT23NOV751 www.ijisrt.com 2671

Volume 8, Issue 11, November 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
selection approach. The RF has extremely high accuracy and  Support Vector Machine
robustness. (Benny et al., 2020)In comparison to other It's a machine learning method that was created mostly
algorithms, such as Logistic Regression (LR), which with binary classification in mind. Supervised learning
achieved accuracy of 87.99%, precision of 89.19%, recall of techniques, or SVMs, help analyze datasets and categorize
93.90%, and f1-score of 0.9148, Qingxiang & Jiesen 2021 them into classes[10] The SVM method works well in large
report that the iterative RF algorithm performed dimensions and is incredibly memory efficient. Among the
exceptionally well in predicting students' performance, with benefits are that it performs best in large dimensions and
an accuracy of 88.55%, precision of 88.21%, recall of functions effectively when there is a distinction between the
95.86%, and f1-score of 0.9187. The general regression classes in the data. It uses the kernel to handle nonlinear
neural network (GRNN) model had the worst performance. data and is reasonably memory efficient. Some drawbacks
GRNN obtained 84.35% accuracy, 86.40% precision, include the inability to function properly in the presence of
90.75% recall, and 0.8852 f1-score. The RF algorithm noise in the dataset, unsuitability for larger datasets, and
performs better in classification when compared to other lengthy training times [10]
algorithms. Large-scale data processing, large-scale variable
parameter support, and intuitive feature importance I. Personalization.
evaluation are all capabilities of this system. Due to its To personalize something is to create something that is
bagging effect, the RF algorithm has demonstrated through tailored to a person's needs. The e-learning platforms ought
numerous algorithm practices and competitions that it has to adjust to various user attributes, including abilities and
superior robustness and stability, as well as good preferences. In order to increase a learner's pleasure and
classification performance[11] success in the course, the functionality allows the system to
specifically address the needs and characteristics of each
individual learner, including their degrees of experience,
prior knowledge, cognitive capacities, skills, interests,
preferences, and learning styles. The goal of the
personalized learning strategy is to give each student an
efficient, effective, and customized learning route so they
may all take part in the process[7].

Fig. 1: Personalisation processing.

 Learning Paths vs Learning Styles these actions might lead to pathways called Learning Paths
The learning route targets a set of objectives by the time (LP). Activities include things like tasks, discussions,
it is finished, but it strives to accomplish one goal at a time. forums, and quizzes. A learning path is intended to give
The information gathered by the learning management students a direct path through a course of study. It simplifies
system about users can be used to define the learner's course material into digestible portions so that students can
profile, understand their behaviour, and pinpoint their quickly assimilate knowledge and move through each course
requirements and challenges. Observing the activities that without difficulty[5]
students take on the system is one method to support them;

IJISRT23NOV751 www.ijisrt.com 2672

Volume 8, Issue 11, November 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig. 2: Path through Learning Style [1]

III. METHODOLOGY with five folds. The clicking position and behavior are key
in the learner’s monitoring activities [8] Pre-processing
Various Machine Learning techniques are used to methods, the dataset, and specifics of the machine learning
study student progress and engagement in different virtual model were used. Overall information was taken into
learning environment activities (VLE). These techniques are account, including demographic information, clickstream
suitable for both domain and categorical educational data for each student every quarter, and student engagement
attributes. The log files in VLE are a typical source of on the VLE portal [15]
information regarding user-system interaction, these records
include every login and activity, and the model was IV. RESEARCH DESIGN
constructed using the information from thelogs.
Weintegrated the model into the Moodle and monitored the Figure 3.Give the explicit process which was carried
learners and validate the model using the cross validation out during the model development.

Fig. 3: Algorithm modeling Process

V. RESULTS AND CONCLUSION learning style pattern that, when the optimal algorithm is
applied to the learning platform, promotes the optimal
E-learning platforms are being utilized more and more learning path in terms of cost and time. Student motivation
to deliver effective learning services, particularly since the and outcomes are significantly impacted by the path analysis
World Health Organization declared the COVID-19 model, individual needs, and the perceived value of e-
pandemic to be worldwide. Institutions of higher learning learning. Furthermore, the perceived value of e-learning
have embraced e-learning. We categorized machine learning outcomes is significantly impacted, albeit indirectly, by
algorithms used in e-learning and helped with student motivation[14]. [15] asserts that early intervention
personalization and adaptation, which helped create a methods for at-risk students should be provided in order to
positive learning environment for the learners. In this paper, identify them. According to the findings, the suggested
we presented a thorough review of the efforts made to apply model can classify data with an accuracy of 84%–93%.The
information and communication technologies to improve e- results demonstrate that deep artificial neural networks
learning services. This paper also explains how to offer perform better than the standard logistic regression and
more flexible and customized course delivery and support vector machine models. The support vector machine
personalized learning by leveraging data collected in big achieves 79.95%–89.14% accuracy, whereas logistic
data-based e-learning systems. Additionally, demonstrated a regression achieves 79.82%–85.60% [14]

IJISRT23NOV751 www.ijisrt.com 2673

Volume 8, Issue 11, November 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
REFERENCES

[1]. Hassan A&El-Sabagh, (2021) Adaptive e-learning

environment based on learning styles and its impact
on the development of students’ engagement.
[2]. Fan Ouyang et al., (2023) Integration of artificial
intelligence performance prediction and learning
analytics to improve student learning in online
engineering course.
[3]. Agung &Wahyu, (2021).Comparison J48 and Naïve
Bayes Methods in Educational Analysis. Comparison
J48 and Naïve Bayes Methods in Educational Analysis
- IOPscience
[4]. Huang, A. Y. Q., Lu, Huang, (2020). Predicting
students’ academic performance by using educational
big data and learning analytics: Evaluation of
classification methods and learning logs. Interact.
Learn.
[5]. David et al., (2021), A New Learning Path Model for
E-Learning Systems. Available from:
https://www.researchgate.net/publication/350559518_
A_New_Learning_Path_Model_for_E-
Learning_Systems [accessed Oct 28 2022].
[6]. Lu, O. H., Huang, A. Y., Huang, J. C. (2018).
Applying learning analytics for the early prediction of
students’ academic performance in blended learning. J.
Educ. Technol. Soc. 21, 220–232.
[7]. Hussein & Al-Chalabi, (2021), Adaptive e-learning
environment based on learning styles and its impact
on the development of students’ engagement.
[8]. Jiang, G., Feng, X., Liu, W., and Xingjun, L. (2020).
Clicking position and user posting behavior in online
review systems: a data-driven agent-based modeling
approach. Inform. Sci. 512, 161–174. doi:
10.1016/j.ins.2019.09.053
[9]. Lassaad K (2022), Using Ensemble Learning
Algorithms to Predict Student Failure and Enabling
Customized Educational Paths.
[10]. Maryam et al (2020). Major depressive disorder
assessment via enhanced k-nearest neighbor method
and signals. Physical and Engineering Sciences in
Medicine, 43(3):1007–1018, 2020.
[11]. Qingxiang & Jiesen (2021) Application of Random
Forest Algorithm in Physical Education.
[12]. Chao Pan, Cheng Shi, Honglang Mu, Jie Li, and Xinbo
Gao. Eeg-based emotion recognition using logistic
regression with gaussian kernel and laplacian prior and
investigation of critical frequency bands. Applied
Sciences, 10(5), 2020.
[13]. Pallavi Asthana & Bramah Hazela (2020) Applications
of Machine Learning in Improving Learning
Environment
[14]. Fethia Yahiaoui et al., (2023), The Impact of e-
Learning Systems on Motivating Students and
Enhancing Their Outcomes during COVID-19: A
Mixed-Method Approach.
[15]. https://www.journalijar.com/upload-articles/

IJISRT23NOV751 www.ijisrt.com 2674