Predicting individual thermal comfort using machine learning algorithms

Human thermal sensation in an environment may be delayed, which may lead to life threatening conditions, such as hypothermia and hyperthermia. This is especially true for senior citizens, as aging alters the thermal perception in humans. We envision a decision support system that predicts human thermal comfort in real-time using various environmental conditions as well psychological and physiological features, and suggest corresponding actions, which can significantly improve overall thermal comfort and health of individuals, especially senior citizens. The key to realize this vision is an accurate thermal comfort model. We propose a novel machine learning based approach to learn an individual's thermal comfort model. This approach identifies the best set of features, and then learns a classifier that takes a feature vector as input and outputs a corresponding thermal sensation class (i.e. "feeling cold", "neutral" and "feeling warm"). Evaluation using a large-scale publicly available data demonstrates that when using Support Vector Machines (SVM) classifiers, the accuracy of our approach is 76.7%, over two times higher than that of the widely adopted Fanger's model (which only achieves accuracy of 35.4%). In addition, our study indicates that two factors, a person's age and outdoor temperature that are not included in Fanger's model, play an important role in thermal comfort, which is a finding interesting in its own right.