An Efficient Bidirectional LSTM-Based Deep Neural Network for Automatic Emotion Recognition Using EEG Signal

Since the variations in brain activity provide a pathway for different emotional states, emotion recognition using electroencephalogram (EEG) has embraced a vast research area in the realm of human-computer interaction. This paper proposes a novel deep neural architecture based on bidirectional long short-term memory (BiLSTM) for automated emotion classification. In the proposed scheme, the long-short-term memory (LSTM) blocks effectively capture important information throughout time steps. The deep-stacked bidirectional mechanism attributes essential features in the forward and backward directions for time-series data. Finally, the acquired feature vector is applied to the dense classifier to categorize different emotion classes. In contrast to the conventional method, an additional feature extraction step is eliminated, resulting in a substantially reduced computational complexity. In this work, extensive and detailed experiments are conducted on a widely available dataset, and satisfactory results are obtained for the valence and arousal domains, considering the performance of all subjects. In binary classification performance, the average accuracy for the valence domain is 82.36%, and the average for the arousal domain is 83.10%.