Development of a D-centred thermal comfort classification model based on the ASHRAE global thermal comfort database II: an Indian case study

The management of thermal comfort in buildings has significant implications for both occupant well-being and energy efficiency. Despite advances in building technology, existing thermal comfort models (such as PMV or single-metric indices) are limited by their reliance on only one aspect of occupant feedback—typically thermal sensation or acceptability—which does not fully represent the complexity of human comfort as observed in real-world data. Analysis of the ASHRAE Global Thermal Comfort Database II reveals correlation coefficients of only 0.35 between predicted mean vote (PMV) and observed thermal sensation, highlighting this disconnect. This paper introduces the D-centred thermal comfort classification (DC3) model, a novel approach that integrates three key subjective indices: thermal sensation, thermal preference, and thermal acceptability. The DC3 model features a systematic 21-class encoding system centred around neutral thermal sensation (‘D’), with each class providing both diagnostic information and prescriptive directional cues for thermal comfort optimisation. Using a dataset of 12,116 thermal comfort observations from India, we develop a random forest (RF) classifier incorporating air temperature, relative humidity, air velocity, and subjective thermal metrics. The model demonstrates 94% accuracy in predicting comfort states, significantly outperforming traditional approaches. The DC3 model’s comprehensive classification framework enables thermal state identification and targeted HVAC control strategies, advancing both personalised comfort delivery and energy-efficient building operation.