Comparison of Human Emotion Classification on Single-Channel and Multi-Channel EEG using Gate Recurrent Unit Algorithm

The use of EEG to recognize human emotions has become a notable trend and breakthrough today. EEG-based emotion recognition is a form of research that uses biomedical signals to distinguish a person's psychological condition (without directly paying attention to changes in facial gestures and attitudes). However, there are many studies related to emotion recognition whose classification accuracy is still low and needs to be improved. Therefore, we propose an EEG-based recognition of positive and negative emotions in this study using the Gate Recurrent Unit (GRU) algorithm. EEG data were taken from 38 participants with four recording channels (FP1, FP2, F7, and F8). In EEG recording, a video was played to stimulate the participants' emotions (positive and negative). Then, the EEG data is processed by filtering, artefact removal, frequency band decomposition, feature extraction, and emotion classification based on signal features. Several classification scenarios (such as by varying the activation function of the classifier and the number of EEG channels) are carried out to obtain an optimal level of accuracy. Based on the emotion classification results (using the Softmax activation function) on multi-channel EEG, the accuracy values reached 98.85% (for training) and 91.45% (for testing).