Statistical Analysis of Multi-channel EEG Signals for Digitizing Human Emotions

The primary objective of this research is the sta-tistical analysis of multi-channel electroencephalogram (EEG) signals for the purpose of emotion recognition performed in the valence-arousal space. The spatial information offered by the sensor location of the multi-channel EEG, is of critical importance as it does not only contain latent information, but also provides insights into the regions of the brain which are active during the expression of the targeted emotions. In particular, the linear correlation between the EEG channel features and the emotion value on the valence-arousal axes is obtained over different frequency ranges using the Pearson method. The five different features utilized in this study are the power of each sensor, power difference between symmetric sensors, power ratio between symmetric differences, average of the sensors readings, and standard deviation of the sensors readings. The statistical analysis was performed using the standard DEAP data set valence, arousal, and dominance values along with raw multi-channel EEG data. Preliminary results indicate that it is possible to optimize the number of sensors used in capturing the EEG signal, while maintaining a high degree of emotion detection accuracy. The standard deviation was found to be the most optimum metric for detecting valence emotion, while the beta frequency range is the better suited for detecting arousal with any of the devised metrics.