An Empirical Evaluation of Methodologies Used for Emotion Recognition via EEG Signals

Abstract

A goal of brain–computer-interface (BCI) research is to accurately classify participants’ emotional status via objective measurements. While there has been a growth in EEG-BCI literature tackling this issue, there exist methodological limitations that undermine its ability to reach conclusions. These include both the nature of the stimuli used to induce emotions and the steps used to process and analyze the data. To highlight and overcome these limitations we appraised whether previous literature using commonly used, widely available, datasets is purportedly classifying between emotions based on emotion-related signals of interest and/or non-emotional artifacts. Subsequently, we propose new methods based on empirically driven, scientifically rigorous, foundations. We close by providing guidance to any researcher involved or wanting to work within this dynamic research field.