Machine learning-based prediction of thermal comfort: exploring building types, climate, ventilation strategies, and seasonal variations

Abstract

Thermal comfort is a critical aspect of building design, directly influencing occupant health, productivity, and well-being. The increasing complexity of modern buildings, coupled with diverse environmental and physiological factors, necessitates advanced approaches to predict and enhance thermal comfort. Unlike prior research, this study applies machine learning with macro-contextual variables such as climate class, season, building type, and ventilation strategy to predict thermal sensation votes. Using the ASHRAE Global Thermal Comfort Database II, over 61,000 observations were preprocessed and analyzed to train eight machine learning models. Among these, the XGBoost model with hyperparameter tuning exhibited the best performance, evaluated through metrics of mean squared error, root mean squared error, and R2. Feature importance analysis revealed that climate class and season were more influential than traditional parameters like indoor air temperature and relative humidity, highlighting study’s innovative focus on macro-contextual variables. The findings emphasize the need to balance general use and specific accuracy when choosing between broad applicability and precision. While the inclusion of macro-contextual variables enhances generalizability, the absence of physiological and real-time behavioural data limits predictive precision. This study demonstrates the potential of machine learning in addressing the limitations of traditional models by integrating macro-contextual variables.