Automatic recognition of pleasant content of odours through ElectroEncephaloGraphic activity analysis

This study presents a machine learning approach applied to ElectroEnchephaloGraphic (EEG) response in a group of subjects when exposed to a controlled olfactory stimulation experiment. In the literature, in fact, there are controversial results on EEG response to odorants. This study proposes a robust leave-one-subject-out classification method to recognize features extracted from EEG signals belonging to pleasant or unpleasant olfactory stimulation classes. An accuracy of 75% has been achieved in a group of 32 subjects. Moreover a set of features extracted from lateral electrodes emphasized that right and left hemispheres behave differently when the subjects are exposed to pleasant or unpleasant odours stimuli.