A PSO weighted ensemble framework with SMOTE balancing for student dropout prediction in smart education systems

Student dropout is a critical issue that affects not only educational institutions but also students' mental well-being, career prospects, and long-term quality of life. The ability to predict dropout rates accurately enables timely interventions that can support students' academic success and psychological resilience. However, the imbalanced nature of student dropout datasets often results in biased and less effective predictive models. To address this, we propose a Particle Swarm Optimization (PSO)-Weighted Ensemble Framework integrated with the Synthetic Minority Oversampling Technique (SMOTE). This methodology balances the dataset using SMOTE, optimizes model hyperparameters, and fine-tunes ensemble weights through PSO to improve predictive performance. The framework achieves 86% accuracy, an AUC score of 0.9593, and enhanced dropout class metrics, including an F1-Score of 0.8633, precision of 0.8633, and recall of 0.86. Compared to Ant Colony Optimization (ACO) and Firefly algorithms, which achieve accuracies of 83% and 85% respectively, our approach demonstrates up to a 3% improvement in key performance metrics. Additionally, in comparison to individual baseline models used in ensemble model Random Forest (RF) with SMOTE (83.5% accuracy, 0.95 AUC) and XGBoost (XGB) with SMOTE (82.7% accuracy, 0.95 AUC), the proposed framework significantly enhances predictive reliability. Furthermore, PSO offers computational efficiency advantages over Firefly and Ant Colony Optimization, reducing hyperparameter tuning time while improving ensemble performance. The proposed framework is scalable and adaptable for real-world applications, particularly in educational institutions, where it can aid in early intervention strategies to mitigate dropout rates.