EEG-Based Emotion Recognition Using Frequency Domain Features and Support Vector Machines

Information about the emotional state of users has become more and more important in human-machine interaction and brain-computer interface. This paper introduces an emotion recognition system based on electroencephalogram (EEG) signals. Experiments using movie elicitation are designed for acquiring subject's EEG signals to classify four emotion states, joy, relax, sad, and fear. After pre-processing the EEG signals, we investigate various kinds of EEG features to build an emotion recognition system. To evaluate classification performance, k-nearest neighbor (kNN) algorithm, multilayer perceptron and support vector machines are used as classifiers. Further, a minimum redundancy-maximum relevance method is used for extracting common critical features across subjects. Experimental results indicate that an average test accuracy of 66.51% for classifying four emotion states can be obtained by using frequency domain features and support vector machines.