Electroencephalogram Emotion Recognition Based on Dispersion Entropy Feature Extraction Using Random Oversampling Imbalanced Data Processing

Abstract

Electroencephalogram (EEG) is the brain’s electrical activity measure, which can reflect people’s inner emotional states objectively. In this article, a dispersion entropy (DispEn) feature extraction-based EEG emotion recognition method is proposed. In feature extraction, the DispEn is computed for the four typical frequency bands, i.e., <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\theta $ </tex-math></inline-formula> , <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\alpha $ </tex-math></inline-formula> , <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\beta $ </tex-math></inline-formula> , and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\gamma $ </tex-math></inline-formula> of EEG signals which are filtered from 32 channels. Furthermore, a random oversampling algorithm is employed to balance the data for the emotional labels to avoid majority biases. The proposed method not only has a better ability to characterize EEG signals but also has a faster recognition speed. In the experiments, the DEAP dataset is used to validate the effectiveness of the proposal, in which the DispEn is extracted from the undecomposed signal and four typical EEG rhythms are compared for emotion recognition by using a support vector machine (SVM). Besides, comparison experiments using DispEn, sample entropy (SampEn), permutation entropy (PerEn), and three other commonly used statistical features are performed. The experimental results show that the proposed method achieves recognition accuracy in high valence (HV)/low valence (LV) and high arousal (HA)/low arousal (LA) is 72.95% and 76.67%, respectively. The computation cost of DispEn feature extraction is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${O}(N)$ </tex-math></inline-formula> that better than some state-of-the-art methods.