Explainable AI Injury Prediction Enhances Outcomes in a School-Based Sports Medicine Football Program for Adolescents

Kaiyue Tang^*Yuelong YeLiuxiang Wei

• School of Physical Education and Health, Guilin Institute of Information Technology, Guilin, China

ABSTRACT Background: Integrating artificial intelligence (AI) with sports medicine in school-based physical education offers new opportunities to enhance adolescent health, improve athletic performance, and reduce injury risk. Objectives: To assess the effects of a 10-week sports medicine–guided football program, augmented by an explainable machine-learning injury-prediction system with individualized feedback, on physical fitness, athletic performance, and injury prevention in junior high school students. Methods: In this prospective cohort study, 195 healthy students participated either in a sports medicine–guided football curriculum incorporating AI-driven feedback or in a standard curriculum. Physical fitness measures—including lung capacity, estimated maximal oxygen uptake (VO₂max), 50-m sprint, standing long jump, and core strength—were collected at baseline and post-intervention. Injury risk was predicted using an explainable ensemble machine-learning model trained on a publicly available football injury dataset and interpreted with SHapley values. Weekly individualized profiles informed training-load adjustments and targeted health-education messages. Results: Students in the intervention group achieved greater gains in cardiopulmonary fitness and lower-limb performance than controls. For males, lung capacity increased from 2485 ± 547 mL to 2698 ± 499 mL (p < 0.01) and VO₂max from 46.9 ± 6.9 to 49.9 ± 4.1 mL·kg⁻¹·min⁻¹ (p < 0.05). For females, lung capacity rose from 2471 ± 722 mL to 2991 ± 632 mL (p < 0.01) and VO₂max from 45.9 ± 3.8 to 47.3 ± 2.5 mL·kg⁻¹·min⁻¹ (p < 0.01). Sprint and jump performances also improved significantly. Reported test performance was MAE = 24.41 days, RMSE = 45.84, R² = 0.734, and Spearman correlation ρ = 0.68, with SHAP analysis highlighting prior injury history and exposure metrics as the most influential features. Conclusion: These findings indicate that integrating sports medicine principles with explainable AI into school PE programs is both feasible and safe, providing individualized risk feedback for adolescents. While direct prevention effects remain to be validated, this approach shows potential for predicted injury risk reduction once tested on age-matched datasets with prospectively monitored injury outcomes.