Classification of Emotion Based on Electroencephalogram Using Convolutional Neural Networks and Recurrent Neural Networks

Abstract

Emotional recognition is widely used in various fields, one of which monitors emotions in understanding the processes that occur in each individual. Electroencephalogram (EEG) can capture emotional information objectively. Nevertheless, it needs appropriate processing. The multichannel of EEG gives much information that yields redundancies. Thus, it can view the information from the multichannel spatially and the sequence between signals as temporal in processing. Several methods are often used, such as Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN). The combination of CNN and RNN utilizes a large amount of information from multi-channel and reserves the characteristics of RNN for processing sequence data such as EEG signals. So that it can maintain the signal sequence information of each channel, this paper proposed the CNN-RNN to identify three positive, negative, and neutral classes of emotions. The EEG signal was filtered at a frequency of 4-45 Hz, according to the signal characteristics of the three emotion classes. It used Wavelet to get Theta, Alpha, Beta, and Gamma waves. The results showed that the use of CNN as multi-channel handling could increase the accuracy, from 91.98%, compared to 79.52% with only RNN. On the other side, CNN-RNN can provide a shorter computation time. However, the choice of emotional duration is significant. Experiments suggest that a second represents more emotional change than five seconds.