EEG-based Happy and Sad Emotions Classification using LSTM and Bidirectional LSTM

Emotion recognition is currently a topic that researchers widely have discussed. This is due to the significant influence of emotions on everyday life. It is not only affecting human health, but also playing an essential role in decision making. Electroencephalogram (EEG) is one of the physiological signals that can be used to measure and recognize emotions based on data from human brain activity. In this study, EEG-based happy and sad emotions classification were performed using time-domain features in the range of EEG bands low alpha(8-10Hz), high alpha(10-13Hz), low beta(13-20Hz), and high beta(20-30Hz) from four different channels namely Fp1, Fp2, F7 and F8 in 10/20 EEG system. The EEG data was obtained from 25 adult subjects. Statistical features such as Mean Absolute Value (MAV), maximum, and variance are used to distinguish between happy and sad emotions. Four classification models were tested and analyzed, namely: logistic regression, SVM, Long-Short Term Memory (LSTM), and bidirectional LSTM. The experiment result showed that the MAV feature contributed to the highest accuracy (95%) using bidirectional LSTM compared to other features. By applying the MAV feature on four different classifiers mentioned above, the highest accuracy was obtained in beta high EEG band.