EEG-based drowsiness detection platform to compare different methodologies

Abstract

Over the years it has been observed that drowsiness appears as one of the factors of the occurrence of driving accidents. By focusing the study on sleep stage 1, transition period between awakeness and sleepiness, it's possible to create a system capable of detecting drowsiness. In this paper, we describe an electroencephalogram (EEG)-based platform capable of detecting drowsiness. This platform consists of the processing and analysis of EEG signals, using several methods to select the most promising features, serving these as input for the creation of different classifiers. Thus, it is possible to study the most appropriate methodology for the development of a prototype capable of detecting drowsiness. The best results were obtained with the use of delays, specifically with 23 and 12, where we used the second and third previous epochs from the past and previous and second previous epochs, respectively, obtaining classifiers with an accuracy of 89.60%, with the delay 23, and 89.45%, with delay 12. Both of these classifiers were SVM with radial basis function kernel.