Abstract:
With the development of computer technology and human-computer interaction technology, the emotion recognition based on EEG has attracted more and more attention. Because researchers often use different stimuli and emotional categories, it is difficult to compare these results in different emotion recognition works. In this paper, we investigate emotion recognition problem via DEAP database. Firstly, we use data clustering technique to determine target classes of human emotional state. And then, we compare two different feature extraction methods, i.e., wavelet transform and nonlinear dynamics, and analyze the effect of baseline features on emotion classification. In addition, we examine five feature reduction algorithms via classification performance and compare the performance of four different classifiers, including K-nearest neighbor, naïve Bayesian, support vector machine, and random forest. It is shown from the results that the combination of kernel spectral regression (KSR) and random forest (RF) can attain the best classifying quality for emotion recognition. Finally, it is found from the study on the relationship between brain regions and emotion that only using a small number of channels, mainly distributed in the frontal cortex, can also achieve a relatively good classification accuracy.