Keywords: Time series analysis, emotional states, neuroscience, connectivity analysis
Identification of the emotional state of a person from electroencephalographic (EEG) data can assist diagnostic and therapeutic procedures for patients with emotional behavior disorders or be useful in more general societal-level inter-person interactions. Herein, comparison of positive and negative valence states (i.e., intrinsic “goodness” and “badness” of a situation) is carried out by means of time series analysis. Multivariate EEG time series and self-rating scores from 32 participants watching 40 selected music videos were obtained from the freely available, online DEAP database and analyzed. Coherence, a linear measure of association in the frequency domain closely related to correlation for bivariate time series, was used to quantify the relations between the 32 time series of each participant and create a coherence matrix with respect to frequency values. Matrices were constructed separately for the two states of valence and for integer frequencies in the range of 4 to 45 Hz. For each participant, state and frequency value, the overall strength of brain connectivity was estimated through the largest eigenvalue (lambda) of the coherence matrix. To compare positive vs negative valance paired t-tests between the lambda values of each participant were performed per frequency. Statistically significant differences (p-value<0.05) in the strength of brain connectivity between the two emotional states were observed in the frequency range of 10 to 18 Hz, which corresponds to the alpha and low beta EEG bands.