Intelligent thermal comfort management in smart buildings using a real-time adaptive RF–XGBoost ensemble model

Many existing thermal comfort prediction models rely on static control logic or single algorithms that adapt poorly to dynamic indoor conditions, creating a gap for adaptive, real-time approaches. This study proposes a hybrid ensemble model that combines Random Forest and XGBoost through dynamic weighting to improve predicted mean vote prediction and HVAC control. Using 52 034 sensor readings from four smart buildings, the model reduces root mean square error by 14% and improves R2 to 0.935 compared with baseline models while maintaining 88.9% accuracy in extreme conditions. Integrated into HVAC operation, it achieves 25.4% energy savings with a 1.81-year payback period and 176% return on investment.