부산대학교 도서관

The motivation of this study was to perceive the gender variations by studying the emotional states (i.e. angry, anxiety, disgust, happiness, sadness and surprise). Emotional electroencephalography (EEG) data were recorded from ten healthful subjects whist the volunteers watched seven, short, emotional audio-visual video clips. Wavelet (WT) has been used as a denoising technique. The spectral relative power ratios ($PR$) features including $(\delta/\theta), (\theta/\alpha), (\alpha/\beta), (\beta/\gamma)$ and ($\theta/\gamma$) were extracted from each EEG channel. In the subsequent step analysis of variance (ANOVA) has been performed to characterize the emotional EEG based on gender differences. Moreover, K-nearest neighbors (KNN) and support vector machine (SVM) classifiers were used to classify the emotional EEG based on gender differences. The results revealed that a relatively high $PR$ for all emotional states in females compared to males particularly in anger, disgust, happiness and surprise emotional states compare to males' $PR$. Moreover, the females show relatively higher $PR$ for anxiety, sadness and neutral in most cases. In contrast, the males show relatively higher $PR$ particularly in $\theta/\alpha$ and $\theta/\gamma$ for anxiety emotional state, higher $\delta/\theta$ and $\alpha/\beta$ for sadness emotional state, and $PR$ particularly had higher $\delta/\theta$ and $\alpha/\beta$ for neutral emotional state. The classification results were 90.4% for SVM and 92% for the KNN. Therefore, the proposed system using WT denoising method, spectral $PR$ features, SVM and KNN classifiers were crucial role in gender identification and characterizing the emotional EEG signals.