Predicting Next Day Heart Rate Variability Based on Training Load in Cyclists Using Machine Learning

Area of Science:

• Sports Science
• Exercise Physiology
• Data Science in Sports

Background:

• Heart rate variability (HRV) is a key indicator of autonomic recovery in endurance athletes.
• Predicting daily HRV fluctuations from training load data is crucial for effective athlete monitoring.
• Current methods lack clarity on the predictive power of common training load metrics for next-day HRV.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in forecasting next-day HRV in competitive cyclists.
• To determine if external load (total mechanical work in kJ) and internal load (session rating of perceived exertion - RPE) improve HRV predictions.
• To compare machine learning approaches (SVR, XGBoost) against a traditional time-series model (ARX).

Main Methods:

• Longitudinal data from seven competitive cyclists over sixteen weeks (590 athlete-days).
• Machine learning models (SVR, XGBoost) and ARX were trained individually per athlete.
• Models were tested using past HRV data only, and with added kJ and RPE, across various lag orders (1-14 days).

Main Results:

• Adding kJ and RPE to HRV-only models yielded only modest improvements in forecasting accuracy (RMSE).
• XGBoost demonstrated the best performance at short lag orders, while model convergence occurred at longer lags.
• Significant individual variability in predictive accuracy highlighted the personalized nature of autonomic responses.

Conclusions:

• Total work (kJ) and RPE provide limited additional predictive information for next-day HRV beyond recent HRV history.
• Broader contextual variables beyond kJ and RPE are likely necessary for substantial improvements in athlete HRV monitoring.
• The individual variability in athlete responses underscores the need for personalized monitoring strategies.