Comparative study on the correlation between human local and overall thermal sensations based on supervised machine learning

Abstract

In heterogeneous indoor environments, significant perceptual discrepancies exist among different body parts concerning their environmental sensitivity. Understanding the relationship between Local Thermal Sensation (LTS) at various body sites and the Overall Thermal Sensation (OTS) is essential for both theoretical inquiry and practical application. Previous studies have predominantly occurred within artificially controlled climatic chambers, with relatively fewer investigations conducted in situ. This study investigates the relationship between LTS and OTS among university students of differing genders in both air-conditioned and non-air-conditioned classroom settings in colder regions. Various supervised machine learning (SML) algorithms were utilized to analyze the data, evaluating their efficacy in predicting the relationship between LTS and OTS and their respective influence weights. The findings demonstrate a significant nonlinear positive correlation between LTS and OTS across different air conditioning settings and genders. Additionally, the Random Forest (RF) algorithm achieved the highest accuracy in predicting LTS weights, with an accuracy exceeding 80%. The study also revealed differences in the influence weights of different body parts across genders and conditions; however, across all conditions, the head and neck region consistently exhibited the highest weight, while the feet displayed the lowest.