Data-driven personal thermal comfort prediction: A literature review

Personal thermal comfort prediction modeling has become a trending topic in efforts to improve individual indoor comfort, a notion that is closely related to the design and performance of building systems, especially in sustainable and smart buildings. This research provides a comprehensive overview of data-driven approaches and processes for predicting personal thermal comfort in a building environment, as derived from a systematic review of 25 studies published in the last 10 years. After refining the concept of personal thermal comfort inspired by predictive modeling in personalized medicine and healthcare, the selection criteria were identified for the reviewed research. Then, three key elements affecting the data-driven modeling process were focused and reviewed, including experimental design, data collection, and modeling techniques. A special emphasis was placed on modeling techniques across the selected studies through a categorization process and comparison of their prediction accuracies. Feature selection and issues important for particular personal thermal comfort models were also reviewed and summarized. Upon reviewing these studies, the authors also considered inter- and intra-individual variability issues in sampling and modeling, data quantity and quality resulting from the collection procedure, model performance, feature importance, and implications for potential online learning techniques. Throughout these analyses, limitations of the current state-of-the-art and possible avenues for future study were addressed.