Enhancing Student Retention and Performance Prediction in E-Learning Environments Using Supervised Machine Learning Models

Attrition rate seems to be a serious problem in the field of online education as it compromises the scalability and efficiency of Virtual Learning Environments (VLEs). This empirical work deals with the issue of student dropout by building a high-dimensional predictive model based on Open University Learning Analytics Dataset (OULAD). Our model uses high-quality supervised machine learning algorithms (i.e. Random Forest, Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost)) to process granular click stream databases and demographic information to predict student performance. In order to address the existing class imbalance in the educational data, the Synthetic Minority Over-sampling Technique (SMOTE) is strictly used. The experimental findings reveal the XGBoost classifier optimized through grid search has a high quality of 92.4% and F1-score of 0.91, which is much better than the baseline models. These findings present a clue to the usefulness of ensemble learning to assist in timely information-based pedagogical interventions.