A Sensing Platform for Personal Comfort Model Development and Application Based on Wearable and Environmental Sensors

Personal Comfort Models (PCMs) are emerging as a data-driven alternative to static indoor thermal standards such as ASHRAE 55 and ISO 7730, which rely on population-averaged setpoints and often fail to satisfy individual occupants. This paper presents WEPOP, a multimodal sensing platform for PCM development, combining a wrist-worn wearable device (capturing electrodermal activity (EDA), photoplethysmographic (PPG), and skin temperature) with room environmental sensors. Using the open-access HEROx dataset (48 participants, campaign conducted in winter), a binary thermal-direction classifier distinguishing Cold from Hot thermal sensation, trained with XGBoost and evaluated under leave-one-subject-out cross-validation, achieves 88.8% accuracy and AUC-ROC of 0.89. An ablation study over three classes (Cold, Neutral, Hot) confirms that fusing wearable physiological signals with environmental measurements improves balanced accuracy from 0.53 (environment only) to 0.60 (full fusion). The Neutral class is not modelled explicitly; instead, a confidence dead-zone in the control system withholds HVAC actuation when model certainty is low, providing an implicit neutral state. The trained models have been integrated into a prototype HVAC control interface with the NEXT.ROOM system, while full occupant-in-the-loop validation at the NEXT.ROOM facility is planned as future work.