Monitoring of cardiac adaptation in elite soccer players over a season through machine learning.

The aim of this study was to assess the evolution of professional soccer players' training status by monitoring an indicator of cardiovascular fitness (ΔHR) over an entire season. The locomotor activity (GPS) and heart rate (HR) of 31 professional soccer players were recorded during small-sided games (SSG) during the 2022-2023 season. Individual predictive models of HR responses built using machine learning methods (i.e. Linear Regression, Support Vector Machine, Random Forest, and eXtreme Gradient Boosting) were trained on a dataset that contains GPS and weather data, Borg CR-10 scale scores and cumulative load. ΔHR was defined as the difference between predicted and measured HR responses. Robustness of models was assessed through a resampling procedure (n = 20). A difference in ΔHR between months was found (p < .05), with a decrease of ΔHR between the early and the middle of the season, and an increase between the middle and the end of the season. The best HR predictive performance was obtained by Random Forest models trained on data including GPS, weather and preceding training load (Mean Absolute Error = 6.59 ± 1.41). Given its ease of use in the context of elite football, ΔHR represents an invisible method to follow elite football players' training status.