Identifying multilayer differential core networks and effective discriminant features for driver fatigue detection

Driver fatigue detection based on EEG has received a lot of attention. In this paper, from the initial brain functional networks of EEG, we first identify multilayer differential core networks based on node and edge discrepancy between different states to capture more discriminant information for the task. Then, the discriminant features are constructed from the resulting multilayer core networks by tensor decomposition and feature selection. The resulting discriminant features contain not only structural information about the network of individual frequency bands, but also additional relational information between the networks of different frequency bands. Our method has the strong ability of information extraction and anti-interference, which can be directly applied to the initial network to identify and utilize weak links with discriminative power. The classification results on the publicly available dataset show that the average accuracy of our method is 9.5% higher than that of the baseline method. When applied to the initial functional network without any pre-processing, the proposed method of this paper can achieve a classification accuracy of 91.14%, which is higher than the results of state-of-the-art related studies on the same dataset. These results indicate that the proposed method is effective and reliable for detecting driver fatigue from EEG.