Predicting Hydration Status Using Machine Learning Models From Physiological and Sweat Biomarkers During Endurance...

Area of Science:

• Exercise Physiology
• Biomarker Analysis
• Machine Learning in Sports Science

Background:

• Athletic performance is significantly impacted by hydration status.
• Noninvasive biomarker data can assess hydration during endurance exercise.
• Previous studies typically involve multiple subjects, limiting personalized insights.

Purpose of the Study:

• To investigate the efficacy of machine learning models in predicting hydration status from single-subject exercise data.
• To evaluate the performance of different machine learning models using physiological and sweat biomarkers.
• To identify key biomarkers for accurate dehydration prediction in personalized monitoring.

Main Methods:

• Conducted 32 constant moderate-intensity exercise sessions on a single subject, with and without fluid intake.
• Measured noninvasive biomarkers: heart rate, core temperature, sweat sodium concentration (from six body regions), and whole-body sweat rate.
• Employed three machine learning models to predict body weight loss (dehydration indicator) and compared prediction accuracy.

Main Results:

• Machine learning models demonstrated similar mean absolute errors in predicting dehydration.
• Nonlinear models generally slightly outperformed linear models.
• Whole-body sweat rate and heart rate provided higher prediction accuracy than core temperature or sweat sodium concentration.
• Sweat sodium concentration from arm patches yielded slightly better accuracy compared to other body regions.

Conclusions:

• Machine learning models can effectively predict hydration status from noninvasive biomarkers in single-subject exercise scenarios.
• Whole-body sweat rate and heart rate are promising indicators for dehydration monitoring.
• This research supports the development of personalized health monitoring systems using wearable sensors and machine learning.