Feature Selection with Sequential Forward Selection Algorithm from Emotion Estimation Based on EEG Signals

Abstract

In this study, we conducted EEG-based emotion recognition on arousal-valence emotion model.We collected our own EEG data with mobile EEG device Emotiv Epoc+ 14 channel by applyingthe visual-aural stimulus. After collection we performed information measurement techniques,statistical methods and time-frequency attribute to obtain key features and created feature space.We wanted to observe the effect of features thus, we performed Sequential Forward Selectionalgorithm to reduce the feature space and compared the performance of accuracies for both allfeatures and diminished features. In the last part, we applied QSVM (Quadratic Support VectorMachines) to classify the features and contrasted the accuracies. We observed that diminishingthe feature space increased our average performance accuracy for arousal-valence dimensionfrom 55% to 65%.