Recognition of emotional states induced by music videos based on nonlinear feature extraction and SOM classification

This research aims at investigating the relationship between Electroencephalogram (EEG) signals and human emotional states. A subject-independent emotion recognition system is proposed using EEG signals collected during emotional audio-visual inductions to classify different classes of continuous valence-arousal model. First, four feature extraction methods based on Approximate Entropy, Spectral entropy, Katz's fractal dimension and Petrosian's fractal dimension were used; then, a two-stage feature selection method based on Dunn index and Sequential forward feature selection algorithm (SFS) algorithm was used to select the most informative feature subsets. Self-Organization Map (SOM) classifier was used to classify different emotional classes with the use of 5-fold cross-validation. The best results were achieved using combination of all features by average accuracies of %68.92 and %71.25 for two classes of valence and arousal, respectively. Furthermore, a hierarchical model which was constructed of two classifiers was used for classifying 4 emotional classes of valence and arousal levels and the average accuracy of %55.15 was achieved.