Classification of Human Emotions using Multiclass Support Vector Machine

Emotion recognition plays a very important role in establishing brain computer interface. Emotion recognition can be done by analyzing speech signal or facial expressions. But these methods cannot be considered as reliable indicators of emotion, because it is possible to generate fake data in these methods. In this paper, Electroencephalography (EEG) is used for detection and classification of different emotions. EEG proves to be more reliable method as it is not possible for the subject to alter the data. The proposed method consists of four steps, viz., data acquisition, pre-processing, feature extraction and classification. Emotions are invoked by using audio visual stimuli. EEG signal is captured for four emotions viz. happy, sad, angry and neutral using power lab instrument by ADInsruments. The recorded EEG signal is then filtered using band pass filter with cutoff frequencies of 3Hz and 30Hz. Discrete Wavelet Transform is applied to the filtered data and then statistical features are extracted. Multiclass Support Vector Machine is incorporated to classify EEG signals into different emotion classes.