Vigilance analysis based on fractal features of EEG signals

Fatigue driving is an important factor leading to fatal traffic accidents. Many different approaches have been studied to detect low level vigilance of drivers. Electroencephalogram (EEG) has been approved an effective medium to measure human vigilance. Fractal dimension (FD) is considered as a useful indicator of the complexity of physiological signal, and maximum fractal length (MFL) is reported to be a practical indicator of the level of human activity. In this paper, we extract above fractal features and linear features from each epoch of EEG data into feature vectors, and then apply Random Forest to the feature reduction and the classification of three different vigilance levels. The result shows that fractal features are more powerful than linear features, the classification accuracy of fractal features reaches 92% on average.