Validation of Wearable Biosignal Sensor-based Estimation of the Physiological Strain Index Using Gaussian Process Regression

Abstract

At physiologically intensive work or during acute exercises, early alert functions are highly required to prevent physiological damage to human health. Wearable sensor-based monitoring of vital parameters can provide real-time measures for the quantification of a worker’s individual psychophysiological and thermal strain to define risk levels for appropriate decision support. One of the most well-recognized indices suitable for use in the workplace so far is the Physiological Strain Index (PSI; Moran et al., 1998) based on sensor data about (i) the core body temperature (CBT) as well as (ii) the heart rate (HR). Until recently, the ground truth information about CBT was particularly measured by cumbersome swallowing expensive gastrointestinal temperature pills. A more comfortable strategy is to attach bioelectrical temperature sensors to the human skin and from these data provide an estimate about the CBT. Dolson et al. (2022) provided a systematic review on distinct algorithms to predict the core body temperature using wearable technology. Most of these algorithms deployed Kalman filters for the prediction. Only a few algorithms incorporated individual and environmental data into their core body temperature prediction, despite the known impact of individual health and situational and environmental factors on the CBT. The presented Machine Learning (ML) framework provides a comparison between a large set of Artificial Intelligence (AI) methods. The Gaussian Process Regression method (GPR; Rasmussen and Williams, 2006) has determined the minimum root mean square error (RMSE) on data from a highly challenging exercise profile applied by a wildland firefighter group. The results are highly competitive with the methods reported in Dolson et al. (2022).

Keywords: Physiological Strain Index, Wearable Biosignal Sensors, Machine Learning

DOI: 10.54941/ahfe1004703