Wearable devices and ML algorithms to assess indoor thermal sensation: metrological analysis

Developing Personal Comfort Models (PCMs) is the best way to manage indoor thermal comfort. This task is not trivial, given subjects’ variability and the plethora of interfering factors affecting the results. Environmental sensors are commonly exploited to this aim, but in recent years also physiological sensors have been investigated. In fact, physiological variables are influenced by environmental conditions, hence they are correlated to the thermal sensation of the subject [1]. Wearable sensors are spreading in different application areas and can be employed also in this context, being user-friendly and of easy application. The potentiality of multi-domain parameters in determining the subject’s thermal sensation (TS), and, hence, thermal comfort can be evaluated. The acquired signals can be ingested by Machine Learning (ML) algorithms with TS classification/prediction purposes [2]. However, the measurement uncertainty in this process should be considered [3], taking into account the diverse sources of uncertainties together with their weights. 2. The aim of this work (pipeline reported in Figure 1) is to show a PCM based on multi-domain physiological signals, as well as to perform its metrological characterization, to identify the variables most affecting the model performance, with a view of optimizing the whole measurement chain.