Hurst exponents based detection of wake-sleep — A pilot study

Detection of Sleep onset is one of complex processes in the area of sleep medicine. The transition from wake state to sleep is termed as sleep onset and is identified using distinct markers like behavioural features, physiological features and changes in EEG Extraction of appropriate features from EEG recordings helps in automated recognition and classification of wake-sleep transition. This research study proposes the introduction of Hurst exponent (HE) to indicate the transition between wake and sleep derived from EEG recordings. Being the non-linear chaotic parameter, Hurst exponent quantifies correlation among the time series data and this property has been exploited for sleep EEGs. Two typical channels O1 and O2 were used for the study and Hurst exponents were estimated for the EEG segments followed by classification using two linear classifiers, LDA and KNN. The statistical analysis confirms that the mean value of HE is lower for sleep than wake. The preliminary study reveals a classification accuracy of 99.96% with HE features with KNN classifier. The procedure needs to be tested with larger datasets.