Reproducible Assessment of Valence and Arousal Based on an EEG Wearable Device

Abstract

An electroencephalography-based detection system of emotional states exploiting few dry channels is proposed. The circumplex model of affect was the reference theory adopted and the standardized dataset International Affective Picture System IAPS was exploited for emotion elicitation. A subset of stimuli polarized on both the valence and the arousal dimension was employed to maximize the effectiveness of the emotion induction. A Self-Assessment Manikin (SAM) was submitted to the subjects after each image to assess the valence and arousal scores of the target emotion. The agreement between the two measures, namely the IAPS scores and the SAM scores was verified through a Bland Altman analysis and a Spearman correlation analysis. An initial screening of the sample allowed to manage the bias caused by depressive and anxiety disorders. The proposed system was experimentally validated. 9 healthy subjects participated in the experimental activity and their EEG signals were acquired through an 8-channel headset. As a result, the best accuracy in the within-subject case of 62.5 ± 4.89 % for the valence dimension and of 66.67 ± 11.88 % for the arousal dimension, was obtained. The poor correlation emerged between IAPS scores and SAM scores negatively impacts on the accuracy and highlights the issue of IAPS update.