Emotion classification based on forehead biosignals using support vector machines in music listening

The purpose of this study was to investigate the feasibility of using forehead biosignals as informative channels for classification of music-induced emotions. Classification of four emotional states in Arousal-Valence space was performed by employing two parallel support vector machines as arousal and valence classifiers. Relative powers of EEG sub-bands, spectral entropy, mean power frequency, and higher order crossings were extracted from each of the three forehead data channels: left Temporalis, Frontalis, and right Temporalis. The inputs of the classifiers were obtained by a feature selection algorithm based on a fuzzy-rough model. The averaged subject-independent classification accuracy of 93.80%, 92.43%, and 86.67% for arousal classification, valence classification, and classification of four emotional states in Arousal-Valence space, respectively, is achieved.