Performance evaluation of personal thermal comfort models for older people based on skin temperature, health perception, behavioural and environmental variables

Abstract

Personal thermal comfort models hold the promise of a more accurate way to predict thermal comfort and therefore a more reliable approach for managing indoor thermal environments. They can be especially relevant as an assistive tool for people with lower thermal sensitivity or with limitations to thermal management and adaptation, such as older people. Nonetheless, although in constant development, studies on personal comfort models continue to focus on office environments and younger adults. This paper explores the development of personal comfort models to predict older people's thermal needs in their homes and evaluates the models' predictive performances in comparison with conventional generalised approaches. Machine learning and environmental, behavioural, health and skin temperature measurements were used to develop individual models for a set of older adults in South Australia. The results show that, on average, the personal thermal comfort models using all studied inputs, except for health perception, presented an optimal accuracy of 66.72%, a Cohen's Kappa of 50.08% and AUC of 0.77, a superior performance when compared with generalised approaches. Results have also highlighted the need for further research on combining physiological sensing, individualised predictive modelling and wearable comfort systems, as well as on defining thermal preference misclassification costs in the context of older people.