Enhancing personal comfort: A machine learning approach using physiological and environmental signals measurements

•Physiological and environmental parameters were collected on 24 volunteers.•Physiological features - thermal sensation (TS) correlation was investigated.•Machine learning algorithms were fed with an optimal features subset to classify TS.•RF classifier provided the best accuracy (90%) for both hot and cold TS. The assessment of the occupants’ thermal sensation (TS) in a living environment is fundamental to enhance well-being and optimize building energy consumption. Machine Learning (ML)-based approaches can be adopted for TS prediction exploiting physiological and environmental parameters, but identifying an optimal features subset is fundamental. This work aims at assessing the correlation between physiological parameters and TS, hence selecting the optimal feature subset for ML-based TS prediction. A dedicated experimental campaign was designed to gather signals through wearable sensors; the actual TS was collected via a specific questionnaire. The results prove the weight of physiological features on the TS determination; ML classifiers achieved an accuracy of up to ≈90% by using physiological and environmental parameters. The strategic potential of personalized comfort systems enables the optimization of both comfort and energy efficiency of a building according to a human-centric approach.