Strengthen EEG-based emotion recognition using firefly integrated optimization algorithm

Emotion recognition is helpful for human to enhance self-awareness and respond appropriately toward the happenings around them. Due to the complexity and diversity of emotions, EEG-based emotion recognition is still a challenging task in pattern recognition. In order to recognize diverse emotions, we propose a novel firefly integrated optimization algorithm (FIOA) in this paper. It can simultaneously accomplish multiple tasks, i.e. the optimal feature selection, parameter setting and classifier selection according to different EEG-based emotion datasets. The FIOA utilizes a ranking probability objection function to guarantee the high accuracy recognition with less features. Moreover, the hybrid encoding expression and the dual updating strategy are developed in the FIOA so as to realize the optimal selection of feature subset and classifier without stagnating in the local optimum. In addition to the public DEAP datasets, we also conducted an EEG-based music emotion experiment involving 20 subjects for the validation of the proposed FIOA. After filtering and segmentation, three categories of features were extracted from every EEG signal. Then FIOA was applied to every subject dataset for two pattern recognition of emotions. The results show that the FIOA can automatically find the optimal features, parameter and classifier for different emotion datasets, which greatly reduces the artificial selection workload. Furthermore, comparing with the binary particle swarm optimization (PSObinary) and the binary firefly (FAbinary), the FIOA can achieve the higher accuracy with less features in the emotion recognition.