User-Independent Classification of Emotions in a Mixed Arousal-Valence Model

In this work we classified EEG features connected with emotions elicited by musical videos. To detect emotions, we used a user-independent approach with data coming from multiple participants in order to test the "peak-end rule". Participant's video ratings were processed to create a mixed valence-arousal labelling. Input features were refined using a combination of feature ranking and data reduction based on intrinsic dimensionality search. Compared to previous literature, our results show that the proposed mixed arousal-valence classification is compatible with previous works applying a distinct arousal or valence classification.