Ensemble Classifier Involving Deep Learning Features for Drowsiness Detection from Electroencephalograph Signals

Driver drowsiness detection and alerting is an important feature of smart vehicles. Prolonged driving without rest can result in accidents. The feature of drowsiness detection and timely alert can prevent accidents and save lives. With ability of low cost portable EEG headsets, mental state analysis using EEG signals has gained research importance. In this work, use of EEG signals for driver drowsiness detection is explored. Raw EEG acquired from the device is processed using empirical mode decomposition for artifact removal. Features are extracted from EEG using three different wavelet transforms and deep learning. A weighted ensemble of machine learning algorithms is proposed to detect drowsiness from the features. The performance of the proposed ensemble classifier is measured against Physionet database and the method is able to provide more than 98% accuracy.