Explainable and imbalance-aware holistic swarm optimization for joint feature selection and random forest learning in educational analytics

Abstract

Accurately predicting students’ academic performance is a crucial component of educational analytics because it enables early detection and informed decision-making. However, existing machine learning methods are sensitive to class imbalance, have poor interpretability, and often rely on fixed feature representations. This paper proposes an imbalance-aware and explicable solution to these issues. An integrated framework of Holistic Swarm Optimization and Random Forest algorithms (HSORFC-FS) is presented to simultaneously optimize model hyperparameters and perform embedded feature selection inside a unified swarm-based search process. Unlike traditional optimization strategies that only focus on accuracy, the proposed framework includes a class-imbalance-aware fitness function that clearly promotes a balanced trade-off between minority-class recognition and overall classification performance. Furthermore, SHapley Additive exPlanations (SHAP) is integrated to provide a global and class-wise understanding of the factors influencing academic performance. HSORFC-FS is evaluated using a real-world student academic performance dataset consisting of 1,194 records in four performance categories: poor, good, very good, and excellent. With a classification accuracy of 80.0% and a weighted F1-score of 0.80, experimental results show that HSORFC-FS performs better than standard Random Forest, bagging, artificial neural networks, k-nearest neighbors, logistic regression, and Naïve Bayes models. The HSORFC-FS framework yields 0.62 for the minority (poor) class F1-score. The proposed framework achieves competitive performance in minority-class recognition while maintaining strong overall classification accuracy and balanced performance across all classes. SHAP analysis states that the most important factors are previous academic performance, attendance rate, and study habits. This explanation helps educational stakeholders to make suitable decisions based on data understanding. In summary, the proposed framework proved that the accuracy, equity, and transparency of academic performance prediction models are enhanced by combining explainable artificial intelligence, imbalance-aware learning, and holistic swarm intelligence.