Sweat Loss Estimation Algorithm for Smartwatches

Abstract

This study presents a newly released algorithm for smartwatches – Sweat loss estimation for running activities. A machine learning model (polynomial Kernel Ridge Regression) is used to estimate the sweat loss in milliliters. A clinical dataset of 748 running tests of 568 people was collected and used for training / validation. The data presents a diversity of factors playing an important role in sweat loss: anthropometric parameters of users, distance, ambient temperature and humidity. The data augmentation technique was implemented. One of the key points of the algorithm is an accelerometer-based model for running distance estimation. The model we developed has a mean absolute percentage error (MAPE) = 7.7% and a coefficient of determination (R2) = 0.95 (at distances in the range of 2–20 km). The performance of the fully automatic sweat loss estimation algorithm provides an average root mean square error (RMSE) = 236 ml; more fundamentally, health-related parameter body weight percentage RMSE (RMSEBWP) = 0.33% and R2 = 0.79. To the best of the authors’ knowledge, the algorithm provides the best performance of any existing solution or described in the literature.

Keywords

• IMU
• PPG
• fitness
• running
• sensors
• skin temperature