Analysis and Research of Automatic Sleep Staging Method with EEG signals Using Random Forest Classifier

Abstract

Automatic sleep staging improves work efficiency and reduces labor costs through it. Thus, an EEG signals-based automatic sleep stages classification method with a random forest classifier is analyzed and studied. Then, two-channelEEG signals of Fpz-Cz and Pz-Oz in the Sleep-EDF database are used for validating the method. With the power spectral density of six characteristic waves in EEG signals as features, a random forest classifier is constructed to recognizefive sleep states (W, Nl, N2, N3, and REM). The effects of different cross-validation methods and classifiers are compared. It performs best when using a 10-fold cross-validation and random forest classifier with the overall classification accuracy, macro-average F1 value, and Kappa coefficient reaching 91.57, 69, and 0.819, respectively. Compared with the existing research, this method is simpler and more effective with better robustness and generalization ability and is suitable for automatic implementation.