Preliminary Study Using Autoencoder for Early Detection of Heat Illness from Heart Rate Variability Obtained with Wearable Device

Detecting a heat illness at an early stage can aid in treating heat illness before it becomes severe. Although it has been reported that heat stress affects the heart rate variability (HRV), it is difficult to obtain a large amount of HRV data before and after the onset of heat illness. In this study, we propose a heat illness detection model using an autoencoder (AE) for anomaly detection based on HRV data under healthy conditions. To test this model, we collected HRV data from participants under several conditions, including workers in a hot environment and amateur marathon runners. With these HRV data, we trained and tested the proposed heat illness detection model. The results indicate that the model achieved an area under the curve of 0.93 for the receiver operating characteristic curve, a sensitivity of 100%, and a false-positive (FP) rate of 3.34 times per hour. This study demonstrates the potential for the early detection of heat illness symptoms when HRV analysis is combined with AE networks; improvements are also needed in the AE model to reduce the incidence of FPs to be a reliable detection system.