Yo-Yo IR1-derived aerobic endurance machine-learning prediction in youth male soccer players using demographics...

This study evaluated the predictive value of demographic information and short-term resting heart rate variability (HRV) for estimating aerobic endurance fitness assessed via the Yo-Yo Intermittent Recovery Level 1 test in youth male soccer players.

A total of 275 players (aged 8-18) were tested during pre-season. Data included players' characteristics, resting heart rate and it's variability in supine and standing positions, and Yo-Yo test results. Short-term (1-5 minutes) RR intervals (RRi) were recorded using the Polar Team 2, preprocessed for artifacts, and analyzed for time-, frequency-, and non-linear HRV features. Frequency-domain metrics were extracted from linearly detrended signals to ensure stationarity. Ten datasets combining player characteristics (age, stature, body mass, BMI, body composition, maturity offset, experience, match status) and HRV were used to train multiple machine learning (ML) regression models. Model performance was evaluated using standard metrics, and SHAP values identified key predictors.

The final sample included 237 players median (range) age and BMI were: 13 (8 - 18) years, 18.8 (13.8 - 29.3) kg/m². The best models, Random Forest using 120-second supine data and Random Forrest using 240-second standing data respectively achieved RMSE 2.89 ml/kg/min (344 m), and MAE 2.27 ml/kg/min (270 m), and RMSE 2.82 ml/kg/min (336 m), and MAE 2.26 ml/kg/min (269 m). Compared to a demographics-only baseline for Random Forest returning RMSE of 3.15 ml/kg/min (375 m) and MAE 2.39 ml/kg/min (285 m), the inclusion of HRV features provided a modest but consistent improvement in predictive accuracy. Age, experience, maturity offset, and non-linear HRV features (symbolic dynamics, Higuchi's Fractal Dimension) were most influential.

ML models combining basic characteristics and short-term HRV showed moderate performance. The incremental improvement attributable to HRV was modest, and these exploratory models are not currently ready to replace Yo-Yo field testing or guide individual training decisions.