An Evolutionary Optimized Variational Mode Decomposition for Emotion Recognition

Emotions provide valuable information regarding habits, health, mental activities, etc. Emotion recognition has attracted wide interests in affective computing, medical, brain-computer interface, and other relevant fields. Emotion recognition from EEG signals has facilitated to impaired peoples to intact with real world. Multi-components EEG signal need to decomposed into set of components to extract the hidden information, empirical selection of decomposition parameters can lead to loss of information due to mode mixing, generation of noisy modes, poor signal synthesis, etc. In this paper, optimized variational mode decomposition is proposed for emotion recognition using single-channel EEG signals. The Eigenvector centrality method (EVCM) is employed for dominant channel selection. An optimum number of modes (K
_opt
) and penalty factor (α
_opt
) are selected adaptively for decomposition of non-stationary EEG signals. Time-domain features are elicited from the modes of EEG signals. The posthoc analysis are employed for significant features selection, these features are used as input to different classifiers for emotion classification. The overall accuracy of 97.24% is achieved with an extreme learning machine classifier for a four emotion classification. The performance of the proposed method is evaluated by five performance parameters. F-1 score, false-positive rate, Mathew's correlation coefficient, and Cohen's Kappa of 0.9454, 0.94%, 92.92%, and 0.9633 is obtained. In comparison with traditional variational mode decomposition and existing state-of-the-art using the same dataset, the proposed method shows improved performance of about 4% and 2%.