r-Principal Subspace for Driver Cognitive State Classification

Using EEG signals, a novel technique for driver cognitive state assessment is presented, analyzed and experimentally verified. The proposed technique depends on the singular value decomposition (SVD) in finding the distributed energy of the EEG data matrix A in the direction of the r-principal subspace. This distribution is unique and sensitive to the changes in the cognitive state of the driver due to external stimuli, so it is used as a set of features for classification. The proposed technique is tested with 42 subjects using 128 EEG channels and the results show significant improvements in terms of accuracy, specificity, sensitivity, and false detection in comparison to other recently proposed techniques.