Quantification of workload and characterisation of key performance factors in elite adolescent female volleyball players using machine learning

This study investigates key performance indicators (KPIs) influencing physical jumping performance in elite-level female volleyball players. This study aims to investigate three hypotheses: (1) the quantification of training load with "Stress Training Response" score is better in explaining and predicting jump performance than classic quantification methods such as mean or sum, (2) high intense exercises stand as primary variables explaining jump performance, and (3) non-linear models are better than linear models to explain and predict jump performance. Nineteen elite-level female volleyball players were monitored over a 190-day season. Various training-related parameters, including workload (external and internal) quantification, player positions, and menstrual cycle data, were collected. Machine learning techniques were employed to analyse and predict jump performance based on these variables. The Random Forest model outperformed other models in describing jump performance (R-squared = 0.64). Key performance indicators identified included workload dynamics, age, and the percentage of intense jumps made during the season. Prediction on a new dataset demonstrated promising results (Mean Absolute Error (MAE) = 4.95 cm (Confidence Interval (CI) 4.56 cm and 5.42 cm), R-squared = 0.55 (CI 0.45 and 0.62)). Findings suggest that intense training prior to performance enhances jump performance, with older players exhibiting superior jumping performance.