Automatic Sleep Staging Based on XGBOOST Physiological Signals

Abstract

The sleep staging can provide a feasible method for sleep medicine treatment, and the artificial sleep staging is becoming outdated, although there is still room for the improvement of accuracy of automatic sleep staging, an automatic sleep staging method is proposed based on XGBOOST and physiological signals. Firstly, the EEG signals and heart rate signals with high availability are selected from a database containing physiological signals, and then the physiological signals are newly sampled and the features are extracted in the time domain, the frequency domain and the nonlinear domain. Secondly, Successive Projections Algorithm (SPA) is applied to select features extracted above, and the redundant features are removed away. Finally, the selected feature sets are put into the XGBOOST model for automatic sleep staging, and the accuracy can reach 92.35%.