<![CDATA[This study addresses the prediction-to-action gap in student performance analytics by proposing an interpretable framework that transforms predictive risk scores into adaptive content recommendations. Rather than only identifying at-risk students, the framework integrates performance prediction, interpretable rule extraction, and decision-support simulation to guide adaptive learning interventions. The study used the Open University Learning Analytics Dataset (OULAD), comprising 6,937 student records after filtering and preprocessing from the original 32,593 records. A Random Forest-based framework was adopted because of its interpretability and rule-extraction capability, although XGBoost achieved slightly higher predictive performance. The framework consists of three components: student performance prediction, interpretable decision rule extraction, and a decision-engine simulation for adaptive content recommendation. The predictive model achieved 87.22% accuracy and an AUC-ROC of 0.932. Rule extraction generated 20 human-readable rules with an average of 2.0 conditions per rule, an interpretability score of 1.000, and 81.6% fidelity to the full Random Forest model. The decision-engine simulation classified students by risk level and produced corresponding adaptive recommendations. An estimated Adaptation Gain metric indicated a potential 53.54% improvement in projected student success rates under conservative simulation assumptions. The proposed framework connects prediction with actionable recommendations to support educational decision-making, although real-world intervention validation remains necessary.]]>
Copyrights © 2026
