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

Outdoor walking supports active mobility and well-being, yet outdoor thermal comfort is increasingly challenged under warmer climate. Existing outdoor thermal comfort indices are primarily developed for steady-state conditions and is limited in representing dynamic walking experiences. This study investigates the use of machine learning to predict thermal sensation during outdoor walking under transient microclimatic conditions. Field experiments were conducted in a high-density public housing estate in Hong Kong under hot and humid weather conditions (mean air temperature: 32.7 °C; mean relative humidity: 68.3%; mean wind speed: 0.50 m/s). Microclimatic parameters, body temperatures, and thermal sensation votes were collected from 25 participants. Four classification models, Random Forest, Support Vector Machine, Extreme Gradient Boosting, and k-Nearest Neighbors, were trained to predict thermal sensation using microclimatic variables and physiological variables separately. The results indicate that models based on physiological measurements outperform those relying solely on microclimate conditions, achieving higher overall predictive accuracy. Feature importance analysis using SHAP reveals that solar radiation and air temperature dominate microclimate-based predictions, while forearm and tympanic temperatures are key physiological predictors. These findings suggest that integrating physiological responses with data-driven models provides a promising pathway for the prediction of outdoor thermal sensation in dynamic walking scenarios.