Binary State Prediction of Sleep or Wakefulness Using EEG and EOG Features

Features extracted from the electrooculogram (EOG) or electroencephalogram (EEG) are able to distinguish between waking and sleeping states of a person with 95.2% and 97.5% accuracy, respectively. Sleep EEG data of eight subjects from the Physionet database have been analyzed using 8th order autoregressive model and the model parameters have been used as features, along with Higuchi fractal dimension and sample entropy. Independently, features derived from EOG such as power in a particular band of frequencies and sum of absolute differences were also used. A combination of all these features obtains the maximum accuracy of 98% using 10-fold cross-validation, employing a support vector machine classifier with radial basis function as kernel. Knowing whether a person is awake or asleep is extremely important before presenting sensory stimuli to the patients with disorders of consciousness as a part of neurorehabilitation.