| Summary: | Investigating gender differences based on emotional changes supports automatic interpretation of human intentions and preferences. This allows emotion applications to respond better to requirements and customize interactions based on affective responses. The electroencephalogram (EEG) is a tool that potentially can be used to detect gender differences. The main purpose of this paper is twofold. Firstly, it aims to use both linear and nonlinear features of EEG signals to identify emotional influences on gender behavior. Secondly, it aims to develop an automatic gender recognition model by employing optimization algorithms to identify the most effective channels for gender identification from emotional-based EEG signals. The EEGs of thirty healthy students from the University of Vienna were recorded while they were watched four short video clips depicting the emotions of anger, happiness, sadness and neutral. In this study, the wavelet transform (WT) de-noising technique, linear spectral mean frequency (<inline-formula> <tex-math notation="LaTeX">$meanF$ </tex-math></inline-formula>) and nonlinear multiscale fuzzy entropy (<inline-formula> <tex-math notation="LaTeX">$MFE$ </tex-math></inline-formula>) features were used. The individual performance of these attributes was statistically examined using analysis of variance (ANOVA) to represent the gender behavior in the brain-emotion in females and males. Then, these two features were fused into a set of hybrid spectral-entropy attributes (<inline-formula> <tex-math notation="LaTeX">$SEA$ </tex-math></inline-formula>). Consequently, optimization algorithms including binary gravitation search algorithm (BGSA) and binary particle swarm optimization (BPSO), were employed to identify the optimal channels for gender classification. Finally, the <inline-formula> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula>-nearest neighbors (<inline-formula> <tex-math notation="LaTeX">$k$ </tex-math></inline-formula>NN) classification technique was used for automatic gender identification of an emotional-based EEG dataset. The results show linear and nonlinear features are remarkable neuromarkers for investigating gender-based differences in emotional states. Moreover, the results show significant enhancement in the overall accuracy of classification achieved by using the BGSA optimization algorithm with the proposed hybrid SEA set when compared to individual features. Therefore, the proposed methods were effective in improving the process of automatic gender recognition from the emotional-based EEG signals.
|
|---|
