Affective Brain-Computer Interface using Frontal Entropy Asymmetry for Human- Machine Interaction

This paper aims to explore the frontal EEG correlates for an affective brain-computer interaction system in a social-life scenario. We first perform a multi-band decomposition on frontal EEG signals using complete ensemble empirical mode decomposition with adaptive noise. Then frontal asymmetric entropy features, including asymmetry of sample entropy and fuzzy entropy, as well as frontal transfer entropy, are proposed for emotion recognition. Those features are further selected by mutual information and ANOVA test, and classified by random forest and gradient boosting classifier. Average accuracies on the DEAP dataset are 71.09% for arousal, 77.34% for valence, on the human-machine affective interaction system are 84.51%, validating the reliability and effectiveness of the proposed features for emotion recognition.