Thermal comfort assessment is a critical component of sustainable building design, influencing occupant wellbeing, energy use, and early-stage design decisions. Conventional comfort indices and static simulation methods provide limited support for predictive, uncertaintyaware, and human-centred evaluation in dynamic design contexts. This paper presents an AI-enhanced Extended Reality (XR) framework for probabilistic, design-time thermal comfort assessment that integrates data-driven machine learning with immersive visualisation. A gradient-boosted decision tree (XGBoost) model is trained on the ASHRAE Global Thermal Comfort Database II to classify occupant thermal states (Cold, Neutral, Hot) using key environmental and personal variables. Rather than producing deterministic outputs, the model generates class probability distributions that explicitly represent the uncertainty inherent in human thermal perception. Model inputs and probabilistic predictions are exported through a lightweight, platform-agnostic data interface and integrated into a Unity-based XR environment. Within XR, predicted comfort states are visualised using intuitive spatial cues, while inputs and probability distributions are displayed through interactive panels to support transparent interpretation and scenario-based exploration. The framework is intentionally designed as a design-time decision-support tool, prioritising interpretability and usability over real-time sensing or closed-loop building control. Results demonstrate stable predictive performance consistent with prior field-based comfort studies and highlight the value of probabilistic representation for uncertainty-aware comfort interpretation. The presented framework illustrates how AI-driven comfort prediction and XR can be combined to enhance explainability, engagement, and human-centred decision support in sustainable building design.
Read more.