A sleep staging model for the sleep environment control based on machine learning

To date, dynamic sleep environment has been attracted the focus of researchers. Owing to the individual difference on sleep phase and thermal comfort, changes in sleep environment should be occupant-centered, and precise regulation of the environment required current sleep stages. However, few studies connected occupants and the environment through physiological signal-based model of sleep staging. Therefore, this study tried to develop a data driven sleep staging model with higher accuracy through sleep experiments collecting information. Raw database was processed and selected efficiently according to the characteristics of physiological signals. Finally, the sleep staging model with an average accuracy of 93.9% was built, and other mean indicators (precision: 82.5%, recall: 83.1%, F1 score: 82.8%) performed well. The features adopted by model were found to come from different brain regions, and the global brain signals were suggested to play an important role in the construction of sleep staging model. Moreover, the computational processing of physiology signals should consider their characteristics, i.e., time domain, frequency domain, time-frequency domain and nonlinear characteristics. The model obtained in this study may deliver a credible reference to advance the research on control of sleep environment.