EEG-based driver drowsiness detection using optimized features and support vector machine

All over the globe, drowsiness is among the mainreasons for road accidents. Hence, well-time detection ofdrowsiness can save the lives of many people. The human brainis one of the best sources for drowsiness detection. To approachand simulate a real environment, this study employs the brainsignals recorded in a driving simulation environment, includinga set of 15 EEG channels associated with six male candidatesdeprived of sleep for almost 21 hours. Feature extraction fromsignals results in 16 features from Fourier transform in thestandard brain band (delta, theta, alpha, and beta) and 16features from a combination of wavelet and Fourier transformshaving 4 features in the standard brain bands or a mixture ofbands. Evolutionary algorithms were utilized to find the mostinformative features. Then, the data of consciousness, transientfrom consciousness to drowsiness, and drowsiness states wereapplied to support vector machine (SVM) with three classes andan RBF kernel, the settings of which were done usinggravitational search algorithm where 11 dominant features witha detection accuracy of %88.81 were obtained.

Keywords:

feature selection , drowsiness detection , EEG signal , SVM