Quantifying women's water polo overhead movement volumes using inertial measurement units and machine learning techniques: a cross-sectional study.

Water polo upper limb external monitoring is not routinely used yet is important in modifying injury risk as shooting volume and intensity are associated with upper limb injury. Monitoring has been performed with video analysis by specialist staff however delays timely feedback-with potential athlete health implications. Inertial measurement units (IMU) are a potential solution. Objective was to investigate IMU and machine learning model performance in classifying overhead movements during coach-led training sessions in elite women's water polo.Athletes wore two IMUs in four training sessions and performed coach-led overhead movements. Video footage was coded using custom software. Vector and magnitude features were computed from IMU data to train five machine learning models. Model performance metrics were calculated.18,186 overhead movements were recorded from 10 athletes over 4 training sessions. Movements were coded under five classifications-swimming, high intensity throwing, low intensity throwing, blocking with ball contact and blocking without ball contact. Resultant and uniaxial acceleration and angular velocity were calculated. Five machine learning models were developed and assessed for performance. Random forest model demonstrated the best performance (accuracy: 0.765, macro-F1 score: 0.504). IMUs and random forest machine learning model have the potential for measuring overhead movements within training sessions.