Sports Injuries Classification Using Machine Learning Models on Biomechanical Data

Early detection and prevention of running-related injuries are fundamental to protecting the health of athletes and optimizing performance outcomes. Despite this importance, early diagnosis remains a significant challenge due to the subtle and often nonspecific nature of initial symptoms, as well as the dependence on subjective clinical judgment. To address these limitations, this work proposes a machine learning-based classification framework aimed at enhancing the identification of injury patterns among runners, thus improving diagnostic accuracy. The proposed method explores five classification models, such as random forest, three different feed-forward back propagation neural networks, support vector machine, K-Nearest Neighbors, and Gaussian naive Bayes on a comprehensive dataset encompassing biomechanical, anthropometric, demographic, and training history variables. The feed-forward back-propagation neural network with 1544 and 772 neurons in the first and second hidden layers was the best model, achieving the highest F1-score of 0.980 and 0.983 in the training and test phases, respectively. Consistent performance in unseen data demonstrated the robust learning capacity of the model and strong generalization in classifying running-related injuries. These results underscore the promise of machine learning approaches in supporting objective and scalable decision-making within sports injury prevention and management.