Advancing urban thermal comfort: adaptive ensemble machine learning models for tropical climates

With the progression of global warming, ensuring thermal comfort in outdoor and semi-outdoor environments is crucial for urban livability, particularly in tropical areas. This study establishes a novel ensemble machine learning (ML) to forecast thermal comfort in tropical urban regions, capable of adaptive regressions that incorporate environmental variables and human behaviors. Field investigations and human sensation datasets including thermal comfort metrics across seasons have been collected. Multiple ML algorithms, entailing logistic regression (LR), neural networks (NN), support vector classifiers (SVC), random forests (RF), gradient boosting (XGBoost), innovative federated ensemble models of NN and RF are utilized to predict thermal acceptance, subjective feelings, and adaptive responses. The new outcomes reveal that the innovative ensemble model, which integrates NN and RF, achieves superior predictive accuracy of 0.57 for Thermal Sensation Vote (TSV) and 0.58 for adaptive response prediction, exceeding that of individual models across all prediction tasks. The results highlight new insights into the capability and robustness of ML-driven adaptive models to improve urban design in the context of climate change. The new insights into these models will assist urban planners and architects in creating sustainable and thermally comfortable public spaces in tropical climates, where increasing temperatures and heat islands significantly affect outdoor usability. The research also indicates potential applications of alternative ensemble techniques to bolstering predictive performance and real-world applications.