A Time-Varying Method for Brain Effective Connectivity Analysis of Emotional EEG Data

Due to individual differences, different subjects have different response to different emotions. Only using the electroencephalogram (EEG) data that can best describe emotions, can the constructed brain networks better reflect the brain connection patterns. This paper proposes a window selection method of time-varying sequence to select critical EEG data that are most relevant to emotions. The generalized orthogonalized partial directed coherence (gOPDC) method is used to construct effective brain networks based on critical EEG data, which can explore the statistical relationship and the information flow between nodes. At the same time, the false connections caused by volume conduction in EEG channels can also be overcome. We extract network features for brain effective connectivity analysis and emotion recognition. The results of the experiments indicate that emotion is mainly related to the frontal region of the brain, and the parietal lobe also plays an important role in low valence emotions. The connection of brain network is the densest in negative emotions and the least in calm and relaxed emotions. The average classification accuracy reaches 78.78% by using KNN classifier on critical EEG data, and the classification results are better than using all data and the intermediate data.