Assessment and prediction of pedestrian thermal comfort through machine learning modelling in tropical urban climate of Nagpur City

Urbanization contributes to the elevated temperatures in built environments, creating Urban Heat Islands (UHI). These UHI intensify heat stress levels, hindering outdoor activities and decreasing walkability. Therefore, paying attention to outdoor space design with thermal comfort in mind becomes crucial. This research aims to develop a machine learning model that assesses and predicts pedestrian thermal comfort (PTC) by training a Random Forest algorithm with field survey data and computed thermal comfort index (modified Physiologically Equivalent Temperature- mPET) from RayMan Pro software. A total of 1364 data sets, including microclimate and street geometry data, were collected from ten streets in Nagpur city over two consecutive years in the May month. Two Random Forest Regressor (RFR) models were developed- one integrating all microclimate and street geometry parameters (Model 1). Another model was trained using the identified important features from microclimate (air temperature and solar radiation) and street geometry parameters (Model 2). The RFR Model demonstrated exceptional predictive capabilities for mPET values and thermal sensation classes. This is substantiated by high correlation coefficients (R² > 0.90 at training phase and R² > 0.80 at prediction phase), low mean squared error values (overall MSE < 1.4), and fitting values of true predictive rates (TPR), and false alarm rates (FAR). This research establishes that only air temperature and solar radiation, along with street geometry parameters, can predict PTC. The developed models have potential applicability in analogous climatic conditions, and they also help identify critical areas of PTC to prioritize mitigation measures, which can ultimately reduce heat stress and UHI impact.