Sleep-awake Classification using EEG Band-power-ratios and Complexity Measures

Single-channel sleep EEG data from eight subjects have been analyzed using Lempel-Ziv complexity measure and alpha-delta and gamma-delta power ratios. Complexity values are consistently high during the waking state for all the subjects and low during the sleep state. Similarly, both the power ratios are high when the subjects are awake and become low during sleep. We have obtained an accuracy of 93.9% in classifying EEG epochs into data corresponding to sleep or awake states. The misclassification is mainly arising from the fact that both the complexity values and the power ratios during the REM sleep state are sometimes comparable to the waking state. By adding another signal such as electromyogram or electrooculogram, one maybe able to minimize the misclassification. The uniqueness of our work is that we have been able to achieve a good accuracy using only one EEG channel, two carefully chosen simple features and a linear classifier.