Valence detection for image stimulated EEG data

Brain-computer interface (BCI) is an emerging communication technique that decodes the user's intention or emotional state. Emotional states usually affect one's brain activity, so use of such emotional states can enhance BCI performance. Emotional data may implicitly have specific informative patterns depending on stimuli such as facial expression, objects, or scenery. In this work, we analyzed 28 emotional datasets and focused particularly on the effect of the perceptual complexity of stimuli. To classify valence (emotional) states and perceptual complexity, we used multidimensional features consisting of the temporal, spatial, and spectral domains. Among them, significantly informative features were extracted by multivariate and univariate analyses. We found that, although the multivariate feature extraction performed valence classification comparable to the univariate one, it was notably more accurate in the classification of perceptual complexity.