Injury Risk Prediction in Rugby League Players with Training Volume Data and Machine Learning

Several studies have used machine learning algorithms to create classification models that predict athletic injury occurrence based on training load. Most existing research focuses on non-contact sports and non-contact injuries, with little consensus over which algorithm or training load features are most effective. This study investigates machine learning algorithms and training load features to predict contact and non-contact injuries in professional Rugby League players. Feature contributions were used to interpret the resulting models and identify which training load features significantly influenced predictions. The results show that the random forest algorithm outperform other machine learning algorithms. Model interpretation revealed that the training load features distance and duration contributed the most to predicting non-contact injuries and collision frequency contributes towards contact injuries.