EEG-based Emotion Classification using Bayesian Belief Networks in Pleasant and Unpleasant Odorant Stimuli

Abstract

In this paper, we proposes an emotion classification method based on Bayesian Belief Networks (BBN) to classify the EEG signals which are induced by the olfactory stimuli. In order to stimulate the olfactory organ, the citralva is used as the pleasant smell and 2-melcaptoethanol is used as the unpleasant smell. We placed the 4-channel EEG electrodes on F3, F4 at frontal lobe and T3, T4 at temporal lobe to acquire EEG signals in according to the standard electrode placement, which is called international 10-20 system. The participants are five high school students (4 male and 1 female) whose ages are from 17 to 18. To extract features from EEG signals, the timefrequency analysis is performed by using the Event-Related Spectral Perturbation (ERSP). The average values of relative power of the frequency in each time domain are used as the features for the BBN classifier. To evaluate the performance of the proposed method, we compared the performance of the BBN and Naïve Bayesian Networks (BN). As a result of the comparison, we confirm that the classification rate of the BBN is increased by approximately 11%.