Connectivity network analysis of EEG signals for detecting deception

Abstract

In this paper, we proposed a novel deception detection (authenticity monitoring) method based on functional connectivity of different brain regions. 12-channels EEG signals were recorded. A mutual information analysis method was used to describe and quantify the connectivity information (correlation) of different regions. Following that, we analysed the statistical difference in the connectivity values between the guilty and innocent groups, and the electrode pairs on which there was statistical difference between two groups were selected. Finally, those connectivity values on selected electrode pairs were combined into the feature vector that was then fed into support vector machine classifier to identify the liars and the truth-telling subjects. Experimental results shows that the classification accuracy of 99.85% is obtained. This study proves that the mutual information method is an effective method of feature extraction for EEG signals, which provides a new way for deception detection based on EEG signals.