Predicting Student Success in Online Learning Using Machine Learning and Explainable AI

With the quick expansion of online education, understanding user engagement and performance metrics has become crucial for improving learning outcomes. This study analyzes a dataset which encompasses comprehensive details about user demographics, engagement levels and course performance in online learning environments using machine learning. The motivation behind this work is to uncover factors influencing course completion rates, engagement patterns and overall learning success by exploring eleven different ML models along with the comparison of three data balancing/sampling and six feature selection techniques. The best of the four data balancing techniques, TOMEK Links and the best of the six feature selection techniques, Chi-square test are preferred based on the model performances examined using evaluation metrics and the time taken by the models. Out of all the models, the five best performing models are LightGBM, XGBoost, Decision Tree (entropy), KNN and ADA Boost. Finally, the explainable AI tools LIME and SHAP are used for model interpretation. Decision Tree with accuracy 1 and highest macro F1 score 0.97 became the best model which executed within 0.37 seconds being the least time.