Wearable EEG-Based Classification of Odor-Induced Emotion

Abstract

Wearable brain sensing and affective brain pro-cessing have recently seen surging interest due to advances in neurotechnologies and rapidly expanding application areas, among which consumer neuroscience, neuroergonomics and dig-ital health. Despite significant progress in understanding olfaction and affective cortical processing, several aspects related to odor-induced emotion remain to be clarified. Among these, are the feasibility of emotion classification using wearable electroen-cephalography (EEG), and the reliability of brain metrics previ-ously proposed in the context of different stimuli in cross-domain emotion recognition. In this study we investigated whether wearable EEG power spectral density (PSD) features can be used to reliably discriminate between odor-induced positive and negative emotions. To this goal, subject-independent trial data has been used within a cross-validation procedure with 3 machine learning algorithms (kNN, linear-SVM, RBF-SVM) to classify the neural response to different odor stimuli. We found that RBF-SVM and PSD features in the delta, theta, alpha and gamma bands yield a high accuracy of 86.1% in classifying positive- and negative-emotion induced by odor stimuli. Moreover, we found that brain metrics relevant for emotion-recognition in the context of other types of stimuli (such as visual) carry discriminative value also in the case of odor-induced emotion.