SP-WOA-XGBoost: A Dual-Task Framework for Student Potential Assessment and Competition Recommendation in Vocational Education

Vocational education faces challenges in accurately assessing student potential and identifying candidates for competition participation due to imbalanced data and diverse learning trajectories. This paper proposes SP-WOA-XGBoost, a dual-task framework integrating Sinusoidal Perturbation-enhanced Whale Optimization (SP-WOA) with XGBoost for student potential assessment and competition recommendation. SP-WOA enhances standard whale optimization through Good Nodes Set initialization, inertia-weighted prey search, sinusoidal spiral updates, and sigmoid-based convergence, enabling efficient hyperparameter tuning. To address class imbalance, we compare SMOTE, SMOTEENN, and class-weighted strategies. Experiments on real-world vocational education data (165 students, 41 courses) show that class-weighted SP-WOA-XGBoost achieves the best trade-off, improving accuracy from 72.7% to 78.8% and minority-class F1-score by 73.1% over baseline. Moreover, the framework generates interpretable Top-N competition recommendations, where 50% of the top-ranked students were confirmed awardees. The results demonstrate that the proposed method bridges predictive analytics and actionable decision support in vocational education, offering a scalable and interpretable solution for data-driven competition talent selection.