Sweat Loss Estimation Solution for Smartwatch

Abstract

This study aimed to develop the new fitness function for wearable devices, namely – Sweat loss estimation during running activity. Machine learning model (polynomial Kernel Ridge Regression) was trained and validated with large and diverse dataset. Totally 568 human subjects participated in 748 running tests. Sweat loss contributing factors such as users’ anthropometric parameters, distance, ambient temperature and humidity were distributed in the wide range of values. The performance of fully automatic sweat loss estimation algorithm provides average root mean square error (RMSE) = 236 ml; more important health-related parameter body weight percentage RMSE (RMSEBWP) = 0.33% and coefficient of determination (R2) = 0.79. To the authors' knowledge the algorithm provides the highest performance among existing solutions or ever described in literature.