Improving the separability of drowsiness and alert EEG signals using analytic form of wavelet transform

Abstract

Drowsiness is a state in which individual decision-making ability is declined. Drowsy driving is one of the key factors for worldwide accidents which claim many lives, causing injuries and permanent disability. Electroencephalogram (EEG) signals are the most common source used for the detection of drowsiness because of changes in neuron behavior. More hidden information can be obtained if it is decomposed into multi-components. Therefore, a flexible analytic wavelet transform (FAWT) based drowsiness detection is explored in this paper. FAWT decomposes a signal in multi-frequency subbands. Multiple features are extracted from these subbands and selected using statistical analysis. Different classification techniques are employed to separate drowsy and alert signals. An accuracy of 95.6%, a sensitivity of 95.2%, a specificity of 98.6%, a precision of 99.1%, and 96.1% of area under the curve are achieved with an extreme learning machine classifier. This shows that our developed method can find more hidden information which is helpful for accurate detection of drowsiness.