Collision detection in rugby union using machine learning: Validation of a machine learning algorithm based on inertial sensor data

This study aimed to develop and validate a supervised machine learning algorithm based on a Random Forest (RF) model to automatically detect collision events in elite rugby union players using microelectromechanical sensor (MEMS) data. Inertial data from triaxial accelerometers and gyroscopes were collected from 36 professional players across four competitive matches. A total of 2,100 collision events were labelled through synchronised video analysis and categorised as tackles, rucks, mauls, or scrums. Data were segmented using 2-s sliding window, and time- and frequency-domain features were extracted. Three matches were used for training and one for testing. Model performance was evaluated using precision, recall, and F1-score. The RF classifier achieved a precision of 88.4%, a recall of 87.2%, and an F1-score of 87.8%, demonstrating robust discrimination between collision and non-collision events. Performance remained consistent across player roles and contact types, indicating strong external validity. These results validate the integration of MEMS-derived data with supervised RF learning as a reliable framework for automated impact detection in professional rugby. This approach offers a scalable tool for workload quantification, and tactical performance assessment, with potential integration for broader performance analysis systems.