Identifying Functional Brain Connectivity Patterns for EEG-Based Emotion Recognition.

Previous studies on EEG-based emotion recognition mainly focus on single-channel analysis, which neglect the functional connectivity between different EEG channels. This paper aims to explore the emotion associated functional brain connectivity patterns among different subjects. We proposed a critical subnetwork selection approach and extracted three topological features (strength, clustering coefficient, and eigenvector centrality) based on the constructed brain connectivity networks. The experimental results of 5-fold cross validation on a public emotion EEG dataset called SEED indicate that the common connectivity patterns associated with different emotions do exist, where the coherence connectivity is significantly higher at frontal site in the alpha, beta and gamma bands for the happy emotion, at parietal and occipital sites in the delta band for the sad emotion, and at frontal site in the delta band for the neutral emotion. In addition, the results demonstrate that the topological features considerably outperform the conventional power spectral density feature, and the decision-level fusion strategy achieves the best classification accuracy of 87.04% and the corresponding improvement of 3.78% in comparison with the state-of-the-art using the differential entropy feature on the same dataset.