On Emotion Recognition Through Dynamic Directed Network and Machine Learning Based on EEG

Information flow in a brain functional network has significant effects on the causality between brain regions and emotion generation. Investigation of such causal relationships under different emotional states is key to reveal the emotion generation mechanisms. In this research, the dynamic directed networks are constructed by transfer entropy method based on DEAP electroencephalogram emotion data set. The information exchanging of the brain network is captured in shortterm time scales. The functional separation and integration ability of brain, information flow and robustness of the network in different emotion states are analyzed. The results found that in the high arousal-high valence and low arousal-high valence states, information separation and integration ability became stronger and the robustness turned high. In the same emotional states, the information flow of the brain regions at all directions varied synchronously. Kinds of machine learning methods combined with the characteristics of the dynamic network are adopted to conduct emotion recognition of testees. Compared with the results of different classifiers in emotion recognition, the classifier built by support vector machine had a higher accuracy. With the accuracy of 96.9% of subject-dependent two-classification, a high precision classifier is successfully achieved in the research which provides an effective method for the future investigation on emotion classification and recognition. (c) 2024 L&H Scientific Publishing, LLC. All rights reserved.