Detection of Emotion from EEG Signal Using Deep Learning: Bi-LSTM and GRU

Emotion states are linked to a wide range of human feelings, thoughts, and actions. They affect our ability to act properly in situations like making decisions, perceiving things, and being smart. So, research on recognizing emotions through emotional signs makes Brain Computer Interface (BCI) systems a better subject for clinical uses and social interactions between people. Electroencephalogram (EEG) signal is an electrophysiological method for brain signal recording. Human emotion arises from the brain, so the detection of emotion can be done by using EEG signals to decide human psychological change during the different emotional states. The aim of this study is to classify human emotion named arousal, valence, dominance and liking. The classification has been done for both binary and multiclass classification from EEG signals. Publicly available "DEAP" dataset is engaged in this work, and EEG recording from 13 channels was used. Fast Fourier Transform (FFT) has been used to separate frequency bands from the EEG signal, and then Power Spectrum Density (PSD) features have been extracted. Two deep learning methods, Bidirectional Long Short-Term Memory (Bi-LSTM) and Gated Recurrent Unit (GRU) have been used for the classification of emotion. We found that the GRU+Bi-LSTM model gives the best result for the binary classification with 96.53% accuracy and Bi-LSTM can predict best accuracy of 92.36% for multiclass classification.