Classification of emotional signals from the DEAP dataset

A Brain Computer Interface (BCI) is a useful instrument to support human communication, frequently implemented by using electroencephalography (EEG). Regarding the used communication paradigm, a very large number of strategies exist and, recently, self-induced emotions have been introduced. However, in general the actual emotion-based BCIs are just binary, since they are capable of recognizing just a single emotion. A crucial node is the introduction of more than a single emotional state for improving the efficiency of a BCI. In order to be used in BCIs, signals from different emotional states have to be collected, recognized and classified. In the present paper, a method for mapping several emotional states was described and tested on EEG signals collected from a publicly available dataset for emotion analysis using physiological signals (DEAP). The proposed method, its experimental protocol, and preliminary numerical results on three different emotional states were presented and discussed. The method, based on multiple binary classification, was capable of optimizing the most discriminative channels and the features combination for each emotional state and of recognizing between several emotional states through a polling system.