Ensemble classifier for driver's fatigue detection based on a single EEG channel

Abstract

Driver's fatigue detection, based on electroencephalography (EEG) signals, is a worthy field of research to study evidence regarding how to exactly pre-warn and avoid casualties nowadays. In this study, an EEG-based system of perfect performance and good stability for evaluating driver's fatigue with only one electrode by ensemble learning method is proposed. Given that EEG signals are unstable and non-linear that using several common entropy measurements to analyse EEG signals is more appropriate including spectral entropy, approximate entropy, sample entropy and fuzzy entropy. In this study, unlike other methods using a single classifier, three ensemble approaches (bagging, random forest and boosting) based on three base classifiers were employed and compared. A driving simulator in this study was used for 12 healthy and adult subjects to perform a continuous simulated driving experiment for 1-2 h. The experimental results show that the proposed method can make use of only one electrode (T6) by gradient boosted DT for driver's fatigue detection, while the average classification accuracy is >94%. The findings of this study indicated that a single EEG channel with optimal ensemble classifier may be a good candidate for usage in the portable system for driver's fatigue detection.