Affect Recognition Using Electroencephalography Features

Abstract

Affect is the psychological display of emotion often described with three principal dimensions: 1) valence 2) arousal and 3) dominance. This thesis work explores the ability of computers to recognize human emotions using Electroencephalography (EEG) features. The development of computer systems to classify human emotions using physiological signals has recently gained pace in the research and technological community. This is because by using EEG to analyze the cognitive state one will be able to establish a direct communication channel between a computer and the human brain. Other applications of recognizing the affective states from EEG include identifying stress and cognitive workload on individuals and assist them in relaxation.

This thesis is an extensive study on the design of paradigms that help computer systems recognize emotional states given a multichannel Electroencephalogram (EEG) segment. The process of first extracting features from the EEG signals using signal processing and then constructing a predictive model via machine learning is often referred to as paradigms. In this work, we will first present a brief review of the state-of-the-art paradigms that have contributed to the topic of emotional affect recognition. Then the proposed paradigms to recognize the principal dimensions of affect are detailed. Feature selection is also performed in order to select the relevant features. The evaluation of the models created to predict the affective states will be performed quantitatively by calculating the generalization accuracy and qualitatively by interpreting them.