ÇAKIR, MUSTAFAAKBULUT, AKHAN2022-09-222022-09-222022Cakir, M., & Akbulut, A. (2022). A bayesian deep neural network approach to seven-point thermal sensation perception. IEEE Access, 10, 5193-5206.https://doi.org/10.1109/ACCESS.2022.3140951https://hdl.handle.net/11413/6769To create and maintain comfortable indoor environments, predicting occupant thermal sensation is an important goal for architects, engineers, and facility managers. The link between thermal comfort, productivity, and health is common knowledge, and researchers have developed many state-of-the-art thermal-sensation models from dozens of research projects over the last 50 years. In addition to these, the use of intelligent data-analysis techniques, such as black-box artificial neural networks (ANNs), is receiving research attention with the aim of designing building thermal-behavior models from collected data. With the convergence of the internet of things (IoT), cloud computing, and artificial intelligence (AI), smart buildings now protect us and keep us comfortable while saving energy and cutting emissions. These types of smart buildings play a vital role in building smart cities of the future. The aim of this study is to help facility managers predict the thermal sensation of the occupants under the given circumstances. To achieve this, we applied a data-driven approach to predict the thermal sensation of occupants of an indoor environment using previously collected data. Our main contribution is to design and evaluate a deep neural network (DNN) for predicting thermal sensations with a high degree of accuracy regardless of building type, climate zone, or a building's heating and/or ventilation methods. We used the second version of the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) Global Thermal Comfort Database to train our model. The hyperparameter-tuning process of the proposed model is optimized using the Bayesian strategy and predicts the thermal sensation of occupants with 78% accuracy, which is much higher than the traditional predicted mean vote (PMV) model and the other shallow and deep networks compared.enArticle