An Efficient Approach to EEG-Based Emotion Recognition using LSTM Network

This work aims to investigate the performance of the Long Short-Term Memory (LSTM) Model for EEG-Based Emotion Recognition. For the experimentation, we use the publicly available DEAP dataset, which consists of preprocessed EEG and physiological signals. Our work limits itself to the study of only the EEG signals to have a scope for developing an efficient headgear model for real-time monitoring of emotions. In this study, we extract the band power, a frequency-domain feature, from the EEG signals and compare the classification accuracies for Valence and Arousal domain for different classifiers. The proposed Long Short-Term Memory (LSTM) model achieves the best classification accuracy of 94.69% and 93.13% for Valence and Arousal scales, respectively, illustrating a significant average increment of 16% in valence and 18% in arousal in comparison to other classifiers.