EEG–Based Emotion Classification Using Convolutional Neural Networks

Abstract

Research in recognizing emotion has been done with many methods, one of them by using brain wave or electroencephalography (EEG). The main benefit of using EEG is because of its ability to track events within brain in millisecond accuracy. Psychologist proposed a model to help classify human emotions, which can be divided into four quadrants using two dimensional emotions of arousal and valence. In this paper, Convolutional Neural Network was proposed as a method for recognizing emotion in the EEG data due to its advantages, such as: local connections, shared weights, pooling and using series of layer to handle high dimensional data of EEG. The overfitting problem in EEG dataset arising from insufficient sample data has been successfully addressed using data augmentation process with effective window size of 4 seconds. The best model is achieved with the accuracy of 72% for arousal and 71% for valence.