Pushing the Boundaries of EEG-based Emotion Classification using Consumer-Grade Wearable Brain-Computer Interfacing Devices and Ensemble Classifiers

Abstract

Emotion classification using features derived from electroencephalography (EEG) is currently one of the major research areas in big data. Although this area of research is not new, the current challenge is now to move from medical-grade EEG acquisition devices to consumer-grade EEG devices. The overwhelmingly large majority of reported studies that have achieved high success rates in such research uses equipment that is beyond the reach of the everyday consumer. Subsequently, EEG-based emotion classification applications, though highly promising and worthwhile to research, largely remain as academic research and not as deployable solutions. In this study, we attempt to use consumer-grade EEG devices commonly referred to as wearable EEG devices that are very economical in cost but have a limited number of sensor electrodes as well as limited signal resolution. Hence, this greatly reduces the number and quality of available EEG signals that can be used as classification features. Additionally, we also attempt to classify into 4 distinct classes as opposed to the more common 2 or 3 class emotion classification task. Moreover, we also additionally attempt to conduct inter-subject classification rather than just intra-subject classification, which again the former is much more challenging than the latter. Using a test cohort of 31 users with stimuli presented via an immersive virtual reality environment, we present results that show that classification accuracies were able to be pushed to beyond 85% using ensemble classification methods in the form of Random Forest.