Comparison among driving state prediction models for car-following condition based on EEG and driving features

Abstract

Risky driving states such as aggressive driving and unstable driving are the cause of many traffic accidents. Many studies have used either driving data or physiological data such as electroencephalography (EEG) to estimate and monitor driving states. However, few studies made comparison among those driving-feature-based, EEG-feature-based and hybrid-feature-based (combination of driving features and EEG features) models. Further, limited types of EEG features have been extracted and investigated in the existing studies. To fill these research gaps aforementioned, this study adopts two EEG analysis techniques (i.e., independent component analysis and brain source localization), two signal processing methods (i.e., power spectrum analysis and wavelets analysis) to extract twelve kinds of EEG features for the short-term driving state prediction. The prediction performance of driving features, EEG features and hybrid features of them was evaluated and compared. The results indicated that EEG-based model has better performance than driving-data-based model (i.e., 83.84% versus 71.59%) and the integrated model of driving features and the full brain regions features extracted by wavelet analysis outperforms other types of features with the highest accuracy of 86.27%.