Evaluating Walking Thermal Comfort in Urban Spaces Using Machine Learning: The Role of Physiological and Microclimate Factors

The recent increase in global temperatures, coupled with ongoing urbanization, has led to more frequent and severe heat waves in urban areas, posing significant health risks to urban populations. This study investigates the prediction of walking thermal comfort using machine learning models built from experimental data. Field measurements and surveys were conducted with 28 participants in hot summer while they walked along a designated route that included open spaces, tree shade, and artificial shade. Three sets of predictors were evaluated using machine learning models: physiological parameters, microclimate conditions, and a combination of microclimate, physiological, and personal factors. Among the physiological models, those incorporating tympanic temperature consistently outperformed models based solely on skin temperatures. In particular, a k-nearest neighbors (KNN) model using calf, chest, and tympanic temperatures achieved the highest accuracy of 73.3%. In contrast, models using only microclimate conditions reached a maximum accuracy of 66.7% with the KNN algorithm, underscoring the stronger predictive power of physiological indicators for walking thermal comfort. Moreover, integrating microclimate, physiological, and personal factors did not yield improved performance over models using physiological parameters alone. These findings highlight the critical role of tympanic temperature in thermal comfort prediction and suggest that focused physiological measurements may offer a more effective approach than broader environmental or combined models. Further research is needed to explore the underlying mechanisms and validate these results across diverse populations and settings.