Audio-visual stimulation based emotion classification by correlated EEG channels

In this paper, a new channel selection technique is presented for emotion classification by electroencephalography (EEG) signals. Audio-visual stimulation is used to generate emotions at the time of experiment. After recording of EEG signals, feature extraction and classification has been applied to classify the emotions (happy, angry, sad and relaxing). The main highlights of the study include: 1) identification/characterization of audio-visual stimulation which generate harmful emotions and 2) proposed approach to reduce the number of EEG channels for emotion classification. Intention behind identification of audio-visual stimulation (video) responsible for harmful emotions like sad and anger is to control their access over social media and another public platform. EEG channels are selected on the basis of their activation probability, calculated from the correlation matrix of EEG channels. Three types of features are extracted from EEG signals, time domain, frequency domain and entropy based. After feature extraction three different algorithms, support vector machine (SVM), artificial neural network (ANN) and naive bayes (NB) are used to classify the emotions. This study is conducted over the DEAP (Database for emotion analysis using Physiological signals) database of EEG signals recorded at different emotional states of several subjects. To compare performance after channel selection, parameters like accuracy, average precision and average recall are calculated. After result analysis, ANN is found as best classifier with 97.74% average accuracy. Among listed features, entropy-based features are found as best features with 90.53% average accuracy.