A Comprehensive Framework for Predicting Student Performance Using Machine Learning in Online Education
Abstract
The rapid expansion of online education has created an urgent need for data-driven tools to predict student performance and enable proactive support. This paper presents a comprehensive machine learning framework designed to model and forecast academic outcomes using behavioral, cognitive, and social metrics. Validated on two large-scale datasets—the Open University Learning Analytics Dataset (OULAD) and Kalboard 360—our framework integrates a rigorous pipeline for data preprocessing, feature engineering, and model interpretation. We engineer novel quantitative features, including a Weekly Engagement Consistency Index (WECI) and an Assignment Timeliness Score (ATS), and evaluate multiple algorithms. The XGBoost model achieved superior performance with an AUC of 0.92 and an F1-Score of 0.88. Crucially, we integrate SHAP (SHapley Additive exPlanations) for model interpretability, revealing WECI and ATS as the most salient predictors. This work demonstrates that a systematic framework leveraging engineered features and ensemble methods can accurately identify at-risk students, providing a scalable solution for early intervention and personalized learning pathways in digital education.