Comparative Analysis of Machine Learning Techniques for Predicting Student Performance in Online Learning Systems

Online learning systems, such as Moodle, have become essential tools for delivering education, especially in scenarios like the COVID-19 pandemic. This study focuses on predicting student performance in such systems using machine learning techniques. We compare various supervised learning algorithms, including Decision Tree, Random Forest, XGBoost, AdaBoost, Support Vector Machines (SVM), LightGBM, Naive Bayes, Neural Networks, and K-Nearest Neighbors (KNN), with a particular emphasis on ensemble learning methods like Random Forest and XGBoost. Our analysis reveals that ensemble methods outperform traditional models, achieving high accuracy in both binary and multiclass classification tasks. Key findings highlight the importance of socio-economic factors, engagement metrics, and demographic characteristics in predicting academic outcomes, with features like imd_band (Index of Multiple Deprivation) and sum_click (student interaction) being the most influential. These insights enable early identification of at-risk students and support the development of personalized learning strategies. The study demonstrates the potential of machine learning to enhance student engagement and retention in online learning environments. Future work will explore hybrid models, real-time data evaluation, and integration with Moodle to create adaptive, automated learning systems. This research contributes to the field of educational data mining by providing actionable insights for improving online education.