KNOWLEDGE EXTRACTION ABOUT SLEEP/WAKE STAGES USING DATA DRIVEN METHODS

This paper describes the process of sleep/wake stage classification. It mainly focuses on the problem of selection of relevant features extracted from the polysomnographic recordings. Iterative features selection methods were applied on a large database composed of several night recordings from different healthy adults. The results showed that the use of relative power of EEG in five frequency bands enables to correctly classify 71% of the whole data base. The addition of features extracted from EOG enables to reach 75% of agreement with the expert classification, but no significant improvement was obtained when adding features extracted from EMG.