Multi-class Emotion Classification Using EEG Signals

Recently, the availability of large EEG datasets, advancements in Brain-Computer interface (BCI) systems and Machine Learning have led to the implementation of deep learning architectures, especially in the analysis of emotions using EEG signals. These signals can be generated by the user while performing various mental, emotional and physical tasks thus, reflecting the brain functionality. Extracting the important feature values from these unprocessed signals remain a vital step in the deployment. Fast Fourier Transformation proves to be better than the traditional feature extraction techniques. In this paper we have compared the deep learning models namely Long Short-term Memory (LSTM) and Convolutional Neural Network (CNN) on 80–20 and 75–25 Train-Test splits. The best result was obtained from LSTM classifier with an accuracy of 88.6% on the liking emotion. CNN also gave a good accuracy of 87.72% due to its capability to extract spatial feature from the input signals. Thus, both these models are quite beneficial in this context.