| Integrative Ensemble Machine Learning Model for Predicting ACL Injury Risk in Adolescent Athletes |
| Paper ID : 1231-SPORTCONGRESS |
| Authors |
|
Sajjad Abdollahi *1, Mehrdad Anbarian2, Rahman Sheikhhoseini3 1PhD Candidate, Department of Sport biomechanics, Faculty of Sports Sciences, Bu-Ali Sina University, Hamedan, Iran. 2Department of Sport biomechanics, Faculty of Sports Sciences, Bu-Ali Sina University, Hamedan, Iran. 3PhD, Associated professor, Department of corrective exercise & Sport injury, Faculty of physical education and sport sciences, Allameh Tabataba'i University, Tehran, Iran. |
| Abstract |
| Introduction: Anterior cruciate ligament (ACL) injuries are among the most prevalent and debilitating conditions affecting adolescent athletes across various sports disciplines. Conventional risk assessment approaches often fail to capture the complex interplay of biomechanical, neuromuscular, and anthropometric factors underlying injury susceptibility. This study aimed to develop and validate an integrative ensemble machine learning model to predict ACL injury risk by combining multidimensional data and improving injury prevention strategies in youth athletic populations. Methods: Data from 468 adolescent athletes (156 injured, 312 controls) were collected through standardized testing protocols covering 25 biomechanical, anthropometric, neuromuscular, and training-related indicators. Six supervised machine learning algorithms—logistic regression, random forest, support vector machine, gradient boosting, k-nearest neighbors, and neural network—were trained and evaluated. An ensemble learning framework integrating the best-performing models was applied to optimize prediction accuracy. Model performance was assessed via cross-validation, and SHAP (Shapley Additive Explanations) analysis was used to interpret variable importance. Results: The ensemble model achieved an accuracy of 87.9%, sensitivity of 85.6%, specificity of 89.3%, and an AUC of 0.923. The most influential predictors included knee valgus angle (0.184), hamstring-to-quadriceps strength ratio (0.156), and ground reaction force (0.142). The SHAP-based interpretability framework provided clinically meaningful insights into individualized risk profiles. Conclusion: The proposed integrative ensemble learning approach significantly outperformed traditional statistical methods, enabling real-time ACL injury risk prediction and personalized prevention planning in adolescent athletes. This framework demonstrates strong potential for integration into athletic monitoring and sports medicine applications. |
| Keywords |
| ACL Injury Prediction, Machine Learning, Ensemble Learning, Sport Biomechanics, Adolescent Football Players, Injury Prevention |
| Status: Abstract Accepted (Oral Presentation) |
