Injury prediction model for lower-limb sports injuries: A novel machine learning-based approach

The complex interaction between various risk factors plays a crucial role in sports injuries. Preventive approaches should consider all of these factors to help identify at-risk individuals. Advanced machine-learning tools and artificial intelligence can be utilized for this purpose. This study aimed to develop an effective injury prediction model for lower-limb sports injuries.

A total of 120 male university-level football, cricket, and basketball players underwent elaborate mental and physical screening, accounting for 44 variables. The injury details of the previous season were recorded. Using these data, different injury-predicting muscle-learning models were built, and their efficiencies were compared.

Thirty-eight injuries were recorded, 20 (52.6 %) of which corresponded to ankle injuries, nine to hamstring injuries (23.6 %), four to knee injuries (10.5 %), and five to foot injuries (13.3 %). Eight injuries occurred during the training phase, and 30 injuries occurred during the competition. Six players sustained the injuries twice. In this case, only the initial injury was considered. Finally, thirty-two injuries were used to construct the model. The model generated by the K Nearest Neighbour algorithm (KNN) with Random Oversampling (ROS) as an oversampling technique was found to be the best (area under the receiver operating characteristic curve score = 0.87, true positive rate = 100 %, true negative rate = 62.5 %).

The machine learning KNN model with ROS showed good accuracy in predicting athletes prone to injuries in football, cricket and basketball. This model would be helpful in injury prevention programs. Larger prospective studies are needed to validate the model.