Predicting daily recovery during long-term endurance training using machine learning analysis.

Area of Science:

• Sports Science
• Data Science
• Physiology

Background:

• Accurate monitoring of athlete recovery is crucial for optimizing training and preventing overtraining.
• Endurance athletes rely on various metrics like training load, sleep, and heart rate variability (HRV) to gauge recovery.
• Predictive modeling offers a potential avenue to personalize recovery insights.

Purpose of the Study:

• To evaluate the efficacy of machine learning models in predicting perceived morning recovery status (AM PRS) and daily heart rate variability (HRV) changes in endurance athletes.
• To identify key variables influencing athlete recovery predictions.

Main Methods:

• Daily monitoring of training, nutrition, sleep, HRV, and well-being in 43 endurance athletes over 12 weeks.
• Construction of global and individualized machine learning models using various algorithms.
• Comparison of model performance against a baseline intercept-only model.

Main Results:

• Group-level models demonstrated lower prediction error (RMSE) for AM PRS and HRV change compared to baseline.
• Individualized models showed improved prediction accuracy over baseline but with significant participant variability.
• Root mean square error (RMSE) for group models: 11.8 (AM PRS) and 0.22 (HRV change).

Conclusions:

• Machine learning can predict daily recovery metrics for endurance athletes at a group level using common data points.
• Individual recovery prediction accuracy varies, indicating that key predictive variables may differ among athletes.
• Further data collection may be necessary to enhance individual-level prediction accuracy.