ABSTRACT
Wearable sensors and machine learning algorithms are widely used for predicting an individual's thermal sensation. However, most of the studies are limited to controlled laboratory experiments with inconvenient wearable sensors without considering the dynamic behavior of ambient conditions. In this study, we focused on predicting individual dynamic thermal sensation based on physiological and psychological data. We designed a smart face mask that can measure skin temperature (SKT) and exhaled breath temperature (EBT) and is powered by a rechargeable battery. Real-time human experiments were performed in a subway cabin with twenty male students under natural conditions. The data were collected using a smartphone application, and we created features using the wavelet decomposition technique. The bagged tree algorithm was selected to train the individual model, which showed an overall accuracy and f-1 score of 98.14% and 96.33%, respectively. An individual's thermal sensation was significantly correlated with SKT, EBT, and associated features.