Sleep stage classification using electroencephalography via Mel frequency cepstral coefficients

In this study, single-channel EEG-based sleep stage classification was investigated using Mel-frequency cepstral coefficients (MFCC), one of the feature extraction methods in the frequency domain. The five-classes classification was performed utilizing a bidirectional long short-term memory network-based classifier using the compressed information of frequency bands (i.e., MFCC). Consequently, it could be found from the F1-score calculated using the best-performed case that the sleep stage could be classified effectively. The highest test accuracy was 92.48%, and the F1-scores of the sleep stages were achieved as 0.8326 (Wake stage), 0.9545 (REM stage), 0.5159(N1 stage), 0.9325 (N2 stage), 0.9500 (N3 stage), respectively.