Machine learning prediction of ACL loading during the wide lunge: a multifactorial coupling analysis based on kinematic and electromyographic signals.

Introduction: In badminton, the wide lunge is a common movement that is highly associated with anterior cruciate ligament (ACL) injury. Existing assessment methods mainly rely on subjective observation, three-dimensional motion capture systems, and surface electromyography signal acquisition, which makes it difficult to rapidly reveal the mechanisms of multi-joint coupling. This study developed and compared six machine learning (ML) algorithms extreme gradient boosting(XGBoost); gradient boosting decision tree (GBDT); random forest (RF); k-nearest neighbours (KNN); kernel ridge regression (KRR); support vector regression (SVR) to predict model-estimated ACL loading during the badminton wide lunge, expressed in multiples of body weight (BW). SHAP was used to rank feature importance and quantify feature contributions. Methods: A total of 237 amateur players were recruited, with kinematic and surface EMG data synchronously collected using a Qualisys motion capture system, AMTI force platforms, and a Delsys surface EMG system during the wide lunge. The sample-size sensitivity and overall predictive performance of the six machine-learning algorithms were evaluated. Results: XGBoost demonstrated the greatest robustness to sample size variation and achieved the best predictive performance ( Conclusion: XGBoost effectively captured the complex non-linear relationships among biomechanical variables. Combined with SHAP-based explainability analysis, insufficient knee flexion, quadriceps-dominant neuromuscular control, and abnormal foot alignment were identified as a combined model-estimated high ACL loading pattern, suggesting that disrupted coupling between multi-joint coordination and muscle activation may be a key mechanism underlying increased injury risk. These findings may provide a basis for individualised training and injury prevention.