Emotion Recognition Using Sparse Graph Analysis of Brain Connectivity

Emotion recognition has gained more importance in recent years due to its various applications in artificial intelligence (AI). Because of high temporal resolution and low acquisition costs, EEG signals have become one of the dominant brain signals for the analysis and recognition of the emotions induced by the nervous system. In this study, we aim to explain brain connectivity using graph models and assess the performance of graph features extracted from brain connectivity. We propose two models to build graphs for brain connectivity and compare their capabilities in expressing the emotional state of a subject. We have used the DEAP dataset for this experiment. Our proposed models show an accuracy increase of approximately 5% compared to solely using brain connectivity features. Another advantage of this approach is that by making the connectivity graph sparser, we could considerably reduce the size of the feature vector compared to the conventional brain connectivity feature vector.