Analysis of Electroencephalographic Signals from a Brain-Computer Interface for Emotions Detection

Despite living in a digital society, the relation between humans and automatic systems is still far from being similar to the interaction among humans. In order to solve the lack of emotional intelligence of those systems, many works have designed algorithms for an automatic recognition of emotions through the assessment of physiological signals, with special interest in electroencephalography (EEG). However, the complexity of professional EEG recording devices limits the possibility to develop and test these algorithms in real life scenarios, out of laboratory conditions. On the contrary, the use of wearable brain-computer interfaces could solve this limitation. For this reason, the present work analyzes EEG signals recorded with a BCI device for the off-line classification of emotional states. Concretely, the spectral power in the different frequency bands of the EEG spectrum has been computed and assessed to discern between high and low levels of valence and arousal. Results reported an interesting classification performance of the BCI device in all frequency bands, being beta waves those which reported the best outcomes, 68.21% of accuracy for valence and 72.54% for arousal. In addition, the application of a sequential forward selection approach before the classification step revealed the relevance of frontal areas for valence detection and posterior regions for arousal identification.