EEG evoked automated emotion recognition using deep convolutional neural network

As life continues to change in the digital era, it is crucial to perceive a person's emotional state. Affective computing is receiving more attention with the increase in the human-computer interface (HCI). Human emotion recognition employing electroen-cephalogram (EEG) signals has been studied to obtain a person's emotional status for different stimuli. However, it is difficult to identify clear patterns in EEG signals because they have low electrical impulses and are highly sensitive to noise. A deep convolutional neural network (DCNN) was employed in the present study to recognize emotions in EEG signals. For this purpose, a publicly available dataset, DREAMER, was utilized in this study to assess the applicability of the model for emotion classification. The dataset consisted of three-dimensional emotions, that is, valence, arousal, and dominance (VAD). 2D emotions arousal and valence were the most-recognized emotions in existing research. The present study identified the 3D emotions present in the above-mentioned dataset. In this study, raw EEG signals from the DREAMER dataset were pre-processed. Subsequently, three EEG rhythms, theta, alpha, and beta, were extracted using a bandpass filter. The power spectral density (PSD) was computed using fast Fourier transform (FFT) in the feature extraction. Finally, a 1D CNN model is applied to the classification of emotions. In addition, the performance of the proposed model was compared with two machine learning (ML) classifiers: random forest (RF) and extreme Gradient Boosting (XGBoost) classifiers. The highest accuracy (ACC) of 97.6% was obtained using the proposed model in the dominance dimension. The working principles were compared and discussed to determine the suitability of the model for emotion recognition applications.