| Abstrakt: |
Electroencephalogram (EEG)-based emotion recognition models are gaining interest as they show the intrinsic state of human. A wide range of features are extracted from the scalp EEG recorded using a different set of electrodes across the brain regions. However, there are no standard set of features accepted amongst researchers for emotion recognition. As a result, new researchers in the field use all features reported in the literature which leads to the curse of dimensionality problem and performance degradation due to high correlation within the feature set. Thus, the primary objective of this work is to improve the performance of the emotion recognition model by using an optimal feature set. This research article proposes differential-evolution-based feature selection (DEFS) algorithm to obtain an optimal feature set for effective subject-independent emotion recognition. The optimal feature set obtained from the DEFS algorithm is used to train the SVM classifier. A wide range of experiments are conducted to analyze the performance of our proposed model using a publicly available EEG-based emotion recognition dataset. The proposed model has been compared with several state-of-the-art feature selection and optimization algorithms. The results are analyzed in the aspects of classification performance, fitness value optimization and computational time. In addition, to assure the subject-independent behavior of the proposed model, subject-wise performance has been analyzed. The proposed DEFS-SVM emotion recognition model has got the classification accuracies of 73.60, 74.23, 71.88 and 71.80% to detect valence arousal, valence, dominance, and liking emotional states, respectively. The experimental results assured that our proposed model outperforms all other algorithms in all aspects. Also, the proposed feature selection algorithm is suitable for any EEG-based emotion recognition model to optimize the feature set. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink