The increasing reliance on data-driven methodologies in education necessitates efficient feature selection techniques to improve student performance prediction. However, existing approaches often struggle to balance feature relevance, redundancy reduction, and computational efficiency. This study proposes EduFeatOpt, a hybrid computational intelligence framework that integrates filter-based methods with metaheuristic optimization. The proposed approach employs Chi-Square Test (CST), ReliefF, and Correlation-Based Feature Selection (CFS) within a Markov Blanket-guided filtering stage (SmartFS) to maximize feature relevance and minimize redundancy. A wrapper-based optimization phase (SmartHive) combines Ant Colony Optimization (ACO) for global exploration and Genetic Algorithm (GA) for local refinement to obtain an optimal feature subset. The model is evaluated on the UCI Student Performance dataset and a real-world College Real Dataset (CRD). Experimental results demonstrate that EduFeatOpt achieves classification accuracies of 91.0%, 85.2%, and 84.5% across datasets using a Support Vector Machine (SVM) classifier. The proposed method improves accuracy by up to 7.11% and reduces feature redundancy by 41.2%, with additional improvements of 10–17% in precision, recall, and F1-score. Statistical validation using ANOVA (p < 0.05) confirms the significance of the results. EduFeatOpt enables scalable and accurate student performance prediction by combining optimizing feature relevance and redundancy, supporting personalized learning and timely academic interventions in high-dimensional educational datasets.
Malik et al. (Thu,) studied this question.