Wavelet analysis for EEG feature extraction in deception detection

Abstract

Deception detection has important clinical and legal implications. However, the reliability of methods for the discrimination between truthful and deceptive responses is still limited. Efforts to improve reliability have examined measures of central nervous system function such as EEG. However, EEG analyses based on either time- or frequency-domain parameters have had mixed results. Because EEG is a nonstationary signal, the use of joint time-frequency features may yield more reliable results for detecting deception. The goal of this study was to investigate the feasibility of deception detection based on EEG features extracted through wavelet transformation. EEG was recorded from 4 electrode sites (F3, F4, F7, F8) during a modified version of the Guilty Knowledge Test (GKT) in 5 subjects. Wavelet analysis revealed significant differences between deceptive and truthful responses. These differences were detected in features whose frequency range roughly corresponds to the EEG beta rhythm and within a time window which coincides with the P300 component. These preliminary results indicate that joint time-frequency EEG features extracted through wavelet analysis may provide a more reliable method for detecting deception than standard ERPs.