Machine Learning for Student Performance Prediction in Online Learning, MOOCS, and Learning Management Systems: A Systematic Literature Review

Abstract

The rapid expansion of online learning in higher education has generated large volumes of learner interaction data through Learning Management Systems (LMSs), Massive Open Online Courses (MOOCs), and related digital platforms. These data provide new opportunities for machine learning to predict academic performance, identify at-risk learners, and support timely intervention.
This study presents a systematic literature review of machine learning approaches used for student performance prediction in online learning environments, with specific focus on MOOCs, LMS data, and digital learning traces. Guided by the PRISMA 2020 framework, the review synthesizes evidence from peer-reviewed studies and addresses five questions: the most common machine learning algorithms, the types of online learning data and predictive features employed, the major prediction targets, the evaluation methods used, and the main research gaps in the field.
The literature indicates that classification-based models dominate the field, with Random Forest, Support Vector Machine, Decision Tree, Artificial Neural Network, and Naïve Bayes among the most frequently used approaches. LMS logs, MOOC clickstreams, assessment records, historical grades, and demographic variables are the most common predictive inputs, while final grades, pass/fail outcomes, dropout, and retention are the main targets.
The review also identifies persistent weaknesses, including limited explainability, weak cross-institutional validation, inconsistent reporting of feature importance, and relatively few studies that evaluate the effect of interventions after prediction. The manuscript concludes with a conceptual framework and a future research agenda to support robust, ethical, and actionable machine learning in online higher education.