A NOVEL HYBRID FEATURE SELECTION FRAMEWORK FOR ENHANCING ACCURACY AND INTERPRETABILITY IN MACHINE LEARNING MODEL FOR STUDENT PERFORMANCE PREDICTION

Abstract

Accurate and interpretable student performance prediction machine learning model are crucial for identifying students at risk and improving academic outcomes in higher education. This study proposes a hybrid feature selection framework integrating Least Absolute Shrinkage and Selection Operator (LASSO) and Ant Colony Optimization (ACO) to enhance predictive accuracy and model interpretability at the same time. Utilizing the original dataset of 35 attributes, the proposed framework able to reduce dimensionality while retaining the most impactful features. LASSO identifies an initial subset of 12 features meanwhile ACO refines to an optimal six feature set in balancing accuracy and interpretability. The Random Forest model was trained on these features and achieved a remarkable accuracy of 99.21% and an AUC-ROC score of 0.9998, outperforming models using the full dataset, solely LASSO-selected features and other features selection techniques. This approach emphasizes the critical role of foundational knowledge, engagement and academic readiness in predicting student success. The proposed framework provides actionable insights for educators, enabling targeted interventions and fostering stakeholder trust in machine learning models for educational data mining