EEG-based multi-band functional connectivity using corrected amplitude envelope correlation for identifying unfavorable driving states.

Recognition of unfavorable driving state (UDS) based on Electroencephalography (EEG) signals and functional connectivity has a significant contribution to reducing casualties. However, when the functional connectivity approach directly applies to recognize drivers' UDS, it may encounter great challenges, because of spurious synchronization phenomenon. We introduce a novel functional connectivity matrix construction approach combined with the ensemble algorithm to identify drivers' UDS in the research. First, EEG data from a previously designed simulated driving experiment containing two driving tasks are extracted, and then functional connectivity matrix construction approach based on amplitude envelope correlation with leakage correction (AEC-c) in multiple frequency bands are calculated. Furthermore, the random subspace is utilized to improve the performances of the k-nearest neighbors (KNN) algorithm. Classification performances of the proposed approach are assessed by confusion matrix, accuracy (ACC), sensitivity (SEN), specificity (SPF), precision (PRE) and receiver operating characteristic (ROC) curve with 5-fold cross-validation strategy. The statistical analysis shows that the regional AEC-c values of 30 EEG channels for the driver's UDS are overall significantly lower than those for the driver's non-unfavorable driving state (NUDS) in the beta, gamma and all frequency bands. Further analysis about performance results shows that the proposed AEC-c-based functional connection matrix analysis approach in all frequency bands combined with the random subspace ensembles KNN achieves a highest ACC of 96.88%. The results suggests that our proposed framework is beneficial for EEG-based driver's UDS recognition, which is helpful to the transmission and interaction of information in man-machine system.