An Automated Approach for Quantifying Emotional Conditions based on Electroencephalogram Signals

Emotion is a psychological process of the mind. It not only has an important role in human interactions, but also has been the subject of attention in human-computer interactions in recent years. Because of diversity and uncertainty nature of the emotion, as well as the individuals’ inability to accurately and quantitatively express their emotions, a quantitative measure for emotion is needed. To describe different emotional states, we used arousal-valence two-dimensional model of emotion.
To propose an approach for emotion recognition based on fuzzy clustering, individuals’ electroencephalogram signals during watching videos along with self-reports about their emotions were achieved from database for emotion analysis using physiological signals (DEAP). Three features of functional connectivity of the brain including correlation, phase locked value, and coherence were investigated and clustered using fuzzy c-means clustering approach. Finally, according to online rating and clustering results for each feature, quantitative and continuous values for valence and arousal were obtained.
Findings: We achieved 0.901 ± 0.079 of valence accuracy and 0.860 ± 0.083 of arousal accuracy. Selected functional connections were related to simultaneous activation of visual, auditory, and sensory perception areas of brain cortex.
The accuracy of the results was more than previous studies which were done on emotion recognition based on binary method. In addition, the results indicated that emotion estimation based on coherence had better accuracy than the other investigated features. This results for valence were more than previous studies. The results of this study are applicable in improving human-computer interactions as well as in the area of cognitive rehabilitation.