Advanced prediction model for individual thermal comfort considering blood glucose and salivary cortisol

Intensified interest in indoor thermal environment has led to an extensive body of research aimed at developing thermal comfort-prediction models with high accuracy. However, previous studies confined the types of bio-signal features due to the limited measuring devices available. Wearable devices for measuring blood glucose (BG) and cortisol (COR) are being developed recently, and the possibility of adding new bio-signal features has been raised. Therefore, this study developed an advanced thermal comfort-prediction model considering BG and salivary cortisol (sCOR) and compared the predictive performance with conventional models. Experiments were conducted to measure the bio-signal features (electrodermal activity, skin temperature, heart rate, blood pressure, BG and sCOR) and psychological measurements of 15 males and 15 females in three conditions: cold, neutral, and warm. To this end, an advanced prediction model was proposed through supervised learning algorithms, including distributed random forest, gradient boosting machine and artificial neural network. The accuracy of the proposed model was 73.4%, yielding 10% better performance than 63.4% of the conventional model. The high feature importance of BG and sCOR demonstrates that these bio-signal features should be included in the prediction model for further studies. The proposed model can be applied in future smart building systems to provide pleasant thermal comfort zones for occupants in general.