Identification and validation of an interpretable machine learning model for predicting peak ACL force during side-cutting maneuvers

To develop an interpretable machine learning model based on biomechanical measurement parameters to predict peak anterior cruciate ligament (ACL) force during side-cutting maneuvers, thereby providing an assessment method for sports training and monitoring. A total of 204 male football and basketball athletes were recruited. Kinematic, kinetic, and electromyographic parameters during side-cutting maneuvers were extracted using the Qualisys optical motion capture system, force plates, and surface electromyography. Machine learning algorithms were used for model training and testing to predict peak ACL force, and Shapley Additive Explanations (SHAP) were employed to interpret feature values. The Extreme Gradient Boosting (XGBoost) model demonstrated high fitting performance (R² = 0.929), with low overall deviation (RMSE = 0.051) and prediction error (MAE = 0.032). Global SHAP importance analysis indicated that knee flexion angle (KFA), hamstring-to-quadriceps activation ratio (H/Q), hip flexion angle (HFA), knee valgus moment (KVM), ankle valgus angle (AVA), hip adduction angle (HAA), and anterior tibial shear force (ASF) were the key features for predicting peak ACL force. The SHAP importance values of KFA (0.07), H/Q (0.05), HFA (0.03), KVM (0.02), and AVA (0.02) showed relatively high contributions, demonstrating strong predictive ability. The increase in peak ACL force during side-cutting maneuvers is essentially the result of the coupling of dynamic postural control mechanical loads and neuromuscular effects. The machine learning model integrating multidimensional indicators achieved high accuracy, improved the predictive capability of peak ACL force in practical applications, and provided an interpretation for the comprehensive analysis of peak ACL force.