Machine Learning in Rugby Union: Predicting and Identifying Key Performance Indicators for Professional Rugby Union...

Area of Science:

• Sports Science
• Biomechanics
• Data Science in Sports

Background:

• Rugby union demands analysis of training and match metrics, with workload's impact on performance indicators understudied.
• Time-motion and video analysis offer insights, but quantitative workload-performance links require further investigation.

Purpose of the Study:

• To investigate the effect of cumulative Global Positioning System (GPS) workload data on key performance indicators (KPIs) in rugby union matches.
• To compare the predictive performance of various machine learning models in understanding workload-KPI relationships.

Main Methods:

• Collected GPS data to calculate 7, 14, and 21-day cumulative workloads.
• Reduced dimensionality using Principal Component Analysis (PCA).
• Employed Linear Regression, Support Vector Regression, Random Forest Regression, and LightGBM to predict KPIs, evaluating models using R², RMSE, and R.

Main Results:

• Individual GPS metrics showed weak correlations with KPIs.
• Machine learning models, especially Random Forest Regression, captured complex, nonlinear workload-KPI interactions.
• Models achieved significant predictive performance (R² 0.40–0.72 for some KPIs).
• SHapley Additive exPlanations (SHAP) enhanced model interpretability, identifying key workload drivers of performance.

Conclusions:

• Machine learning models provide superior prediction of rugby KPIs from workload data compared to simple correlations.
• Understanding workload-performance dynamics through interpretable ML is crucial for effective player management.
• Findings offer actionable insights for optimizing training loads to enhance player performance in rugby union.