Application of human-centric digital twins: Predicting outdoor thermal comfort distribution in Singapore using multi-source data and machine learning

In the face of global climate warming, outdoor thermal comfort in urban settings is increasingly critical. However, accurately predicting residents' thermal perceptions during outdoor activities remains challenging due to complex environmental dynamics. This study introduces a human-centered digital twin framework that integrates physiological data, atmospheric conditions, and urban building environment features, with multiple machine learning models employed to predict and analyze outdoor thermal comfort in different regions of Singapore. Among these methods, the Bayesian-tuned XGBoost model exhibits the highest accuracy (0.66), notably excelling in categorizing “Prefer cooler” and “Prefer no change” responses. SHAP value analysis identifies key influencing factors such as human activity intensity (heart rate), geographical location (longitude and latitude), meteorological conditions (solar azimuth angle, dew point temperature), and greenery (Normalized Difference Vegetation Index). Based on the most effective machine learning method , this research develops a user-personalized real-time prediction model for urban thermal comfort perception. The extensive hourly grid-based prediction results illustrate the spatiotemporal variations in outdoor thermal comfort, highlighting preference differences across locations, seasons, and activity levels. Results underscore the efficacy of the human-centric digital twin approach and machine learning in managing urban thermal environments , leveraging multi-source data to complement traditional survey methods effectively. • XGBoost model achieves a 66 % accuracy rate in predicting outdoor thermal comfort. • Multi-source data reduces reliance on extensive field data collection for thermal comfort predictions • SHAP analysis elucidates the complex nonlinear relationships between various factors and thermal comfort. • Activity intensity is the most influential factor in determining outdoor thermal comfort in Singapore. • Machine learning can help identify thermal discomfort areas, but performance may be limited in data-sparse regions.