Statistical algorithms for emotion classification via functional connectivity

Pattern recognition algorithms decode emotional brain states by using functional connectivity measures which are extracted from EEG signals as input to the statistical classifiers. An open-access EEG dataset for emotional state analysis is used to classify two dominant emotional models, based on valence and arousal. To calculate the functional connectivity between all available pairs of EEG electrodes four different measures, including Pearson's correlation coefficient, phase-locking value, mutual information, and magnitude square coherence estimation, were used. Three kinds of classifiers were applied to categorize single trials into two emotional states in each emotional model (high/low arousal, high/low valence). This procedure resulted in decoding performance of 68.30% and 60.33% for valence and arousal respectively in test trials which were significantly higher than chance (≈ 50%, t-test, and significance level of 0.05). The results obtained using a phase-locking value approach were significantly better than previous findings on the same data set. These results illustrate that functional connectivity between distinct neural populations can be considered as a neural coding mechanism for intrinsic emotional states.