The Predicting Middle School Students’Fitness Scores Using Physical Activity Promotion System (PAPS) Data: An XGBoost Regression Model with SHAP-Based Explainable Analysis

This study aimed to predict middle school students’ total fitness scores using PAPS data and explainable AI. Data were collected from 292 students (150 males, 142 females) across five fitness domains: cardiovascular endurance, strength, flexibility, power, and BMI. An XGBoost regression model combined with SHAP was applied to predict scores and interpret variable importance. The model showed high predictive performance (RMSE = 4.95, R² = 0.915; 5-fold R² = 0.920). SHAP results identified push-ups, BMI, and flexibility as the top contributors. High push-up counts and flexibility had strong positive effects, while BMI above 22 ㎏/㎡ negatively impacted scores. Partial dependence plots revealed non-linear patterns and thresholds. Multicollinearity and residual tests confirmed model stability. The findings demonstrate that explainable machine learning can support personalized feedback and health strategies in school-based physical education.