Supervised machine learning of thermal comfort under different indoor temperatures using EEG measurements

In this paper, machine learning techniques in conjunction with passive EEG (electroencephalogram) measurement were explored to classify occupants’ real-time thermal comfort states, which have the potential in the future for energy saving through adopting time varying set points when real-time changes in thermal comfort can require less energy input. The performances of different machine learning techniques were compared, and the method to select linear continuous features for class interpolation was also explored. For the full-set features, the performances of different classifiers were satisfactory, with classification rates all above 90%. The LDA classifier had the best performance. The second best was the NB classifier, and the relatively worst was the KNN classifier. The linear continuous EEG features were selected by interpolation and can be found for all human subjects. Higher selection threshold led to less selected features but higher average performance of these features. In general, the EEG based machine learning methods can classify occupants’ real-time thermal comfort states, and could potentially lead to more building energy saving through comfort-driven time varying set points.