Discrimination between genuine and acted expressions using EEG signals and machine learning

Abstract

The main purpose of this thesis work was to quantify happiness in an objective manner. This is in line with the objectives of the National Program for Happiness and Positivity in the UAE. The major contribution to this thesis work included designing and conducting experiments to study the emotion-related cognitive process using EEG signals. The focus is to develop a novel method for classifying EEG signals related to genuine and acted expressions. A framework for quantifying three different affective states: actual/true positive, acted/fake and neutral positive emotions were developed. The major stages involved the development of an emotion related EEG database comprising of 28 subjects, feature extraction, and finally the application of machine learning algorithms. Two main approaches were used for feature extraction: the first method included discrete wavelet transform while, the second method involved a combination of discrete wavelet transform (DWT) and empirical mode decomposition (EMD). Average power features extracted from both the techniques were used for classification of the three affective states. Highest accuracy of 69.2 % using the DWT method and 94.2 % using DWT-EMD method was achieved.