Abstract

Personal Thermal Comfort (PTC) is a critical indicator of health, well-being, and productivity, especially for susceptible occupants (e.g., older people and patients). However, PTC prediction is challenging because it requires advanced technologies to deal with complex factors produced by dynamic environments (e.g., ambient factors) and individual preferences (e.g., physiological responses). This research addresses these concerns by designing and developing Deep Neural Network (DNN) model based on occupants' physiological responses, including skin temperature (ST), heart rate (HR), electrodermal activity (EDA), and airflow (AF). The experimental chamber and measurement procedure are proposed to observe physiological signals under the control of indoor temperature and humidity, particle matter concentration (PM<inf>2.5</inf> and PM<inf>10</inf>), and CO<inf>2</inf> concentration. The results show that our DNN model can perform predictive effectiveness of PTC satisfaction levels more effectively than the principle model, achieving approximately 90 % of average precision, recall, and f-measure, improving almost 10 % of rare events. This ensures that the DNN model is a natural fit for predictive individual satisfaction and can be employed in intelligent applications.