Predictive Analytics in Mobile Education: Evaluating Logistic Regression, Random Forest, and Gradient Boosting for Course Completion Forecasting

Abstract

This study aimed to compare the effectiveness of three predictive algorithms—logistic regression, random forest, and GBM—in predicting course completion using user engagement data from online learning platforms. By analyzing engagement metrics such as session duration, session frequency, and quiz scores, the study sought to identify the most effective model for forecasting course completion, providing insights into which aspects of student behavior were most predictive of success. Logistic regression emerged as the best overall performer, achieving the highest accuracy (52.13%) and F1-Score (56.17%), indicating its balanced approach to predicting course completion and non-completion. Random forest and gradient boosting machines (GBM) showed strengths in specific areas; random forest maintained a good balance between precision and recall, while GBM excelled in recall, identifying students likely to complete courses but with lower precision, leading to more false positives. The findings have practical implications for educational technology, particularly in designing personalized learning paths and targeted interventions to support at-risk students. The study also acknowledged limitations, including the dataset’s focus on engagement metrics without demographic context and the potential for model-specific biases. Future research should explore additional predictive features, larger datasets, and more advanced algorithms to enhance the robustness and applicability of predictive models in real-time educational settings.