Characterizing fatigue-related neural signatures via multi-band EEG features and a three-stage feature optimization strategy for driver fatigue detection

Abstract Fatigue driving is one of the primary factors that contribute to road traffic accidents worldwide and poses a serious threat to transportation safety as well as to public life and property. To achieve accurate and efficient detection of driver fatigue while the complex and nonlinear characteristics of physiological signals are adequately represented, this study proposes a detection framework that is based on multi-band electroencephalography (EEG) feature optimization and a support vector machine (SVM). The EEG signals are decomposed into the θ (4–8 Hz), α (8–14 Hz), β (14–30 Hz), and full-band (1–50 Hz) components, from which time-domain, frequency-domain, nonlinear dynamic, and brain network connectivity features are extracted so that a high-dimensional feature space with multi-scale and multi-structural information can be constructed. To reduce redundancy and enhance discriminative capability, a three-stage feature optimization strategy is adopted, in which variance inflation factor analysis is applied to remove multicollinearity, mutual information is used to identify features with strong discriminative power, and SVM-based recursive feature elimination SVM is employed to further compress the feature space. Fatigue states are then identified by SVM classifiers that are constructed with different kernel functions, and the detection performance across frequency bands is systematically evaluated. Experiments are conducted on a publicly available driving EEG dataset, and the results demonstrate that the β -band features combined with a medium Gaussian kernel SVM achieve the best performance. Under cross-validation, the proposed method attains an accuracy of 86.34%, a precision of 88.89%, a sensitivity of 83.06%, an F 1-score of 85.87%, and a false detection rate of 10.38%. These results indicate that the proposed framework effectively captures the complex EEG patterns associated with driver fatigue and provides a robust solution for fatigue detection.