Prediction of Assessment Results Based on XGBoost-SVC Model

This study proposes an integrated hierarchical model combining XGBoost and Support Vector Classifier (SVC) for high-precision student performance prediction and training effectiveness evaluation. By leveraging feature selection, hierarchical training, and dynamic optimization, the model classifies academic performance and quantifies training effectiveness. U sing standardized educational data, key indicators were identified through correlation analysis (e.g., a 0.737 correlation between admission and departure skills) and XGBoost feature importance. In the first stage, models like XGBoost and Random Forest predict departure scores, while SVC is employed for classification in the second stage. Experimental results show that the SVC model achieves an Fl-weighted average of 0.932 in hierarchical cross-validation, confirming prediction robustness. Training effectiveness is quantified by the difference between admission and departure scores, revealing significant improvement in type 4 schools (0.272). The top 5 skill-based ranked schools provide insights for student selection. Sensitivity analysis further confirms the model's generalization ability. This study presents a high-precision, interpretable hybrid modeling approach for educational data mining, combining XGBoost feature selection and SVC boundary optimization.