Predicting education building occupants’ thermal sensation through CatBoost-DF algorithm

Abstract

A novel machine learning method, named CatBoost-DF (CatBoost deep forest), is proposed to solve this existing problem of low accuracy and lack of practicality in thermal sensation prediction. In the CatBoost-DF, a cascading strategy is introduced to strengthen the association between each layer of CatBoost. To verify the accuracy and robustness of CatBoost-DF, experiments collected physiological and environmental data from hundreds of subjects with the help of sensor devices and questionnaires. Compared with existing state-of-the-art machine learning methods, CatBoost-DF shows significant superiority, with a prediction accuracy of 90%, which is 4%-39% higher than other models. Moreover, the study explored the effects of seasonal and gender factors on thermal sensation. Result shown that different seasons have different thermal sensation for males and females. Finally, CatBoost-DF is applied to predict occupants’ thermal sensation, and the “comfort range” of the important parameters HR, WS, and CTR that affect the thermal sensation is calculated experimentally.