<![CDATA[Early identification of academically at risk students is essential for timely intervention and improved retention in higher education. This study investigates the effectiveness of using pre admission and early semester LMS data to predict student risk using machine learning models. The objective is to assess whether limited, readily available data from the first four weeks of instruction can reliably support early warning systems. A supervised learning framework was applied using the Open University Learning Analytics Dataset (OULAD), with features derived from student demographics and early LMS activity logs. Models evaluated include Logistic Regression, XGBoost, and CatBoost, with time based validation and SMOTE employed to address class imbalance. Model performance was measured using ROC AUC, F1 Score, and Recall. The CatBoost model achieved the best performance, with an F1 score of 0.770 and ROC AUC of 0.750, significantly outperforming baseline models. Quiz submission behavior, login frequency, and pre admission qualification level emerged as the most predictive features. Results also revealed a steady week by week improvement in model accuracy, confirming the increasing value of LMS engagement data over time. These findings affirm that early stage student data can be used effectively to predict academic risk, enabling institutions to act before major assessments are conducted. The study emphasizes the need for institutional readiness, ethical implementation, and inclusive practices in deploying predictive tools. Future research should expand the feature space and test cross institutional generalizability to refine early warning systems further.]]>
Copyrights © 2025
