Acquisition and Analysis of EEG Data for Forensic Application

Abstract

Countries throughout the world are increasingly spending on scientific research and technology to improve its security personnel to protect their borders and citizens from any form of illegal activities. The goal of this study was to collect and analyse EEG data for the forensic application of hidden information recognition in the human brain. First of all, a new mock crime protocol was drafted, derived from which, pictorial (images) and auditory (sounds) stimuli were prepared for introduction to all the participants. EEG waveforms of 35 participants were recorded. Following that, features derived from time, frequency, wavelet, and empirical mode decomposition (EMD) were evaluated. Using the ReliefF ranking technique, the top three ranked pictorial and top seven ranked auditory features were produced. On transient waveforms such as EEG, EMD has a high data-adaptive ability. The first contribution of this work was the analysis of EMD-derived features in this EEG data-based concealed details identification investigation. The features were segregated into the two output categories (guilty and innocent) using different classifiers namely artificial neural network (ANN), support vector machine (SVM) (radial basis function (RBF) kernel), SVM (polynomial kernel), and k-nearest neighbor (KNN), utilizing tenfold cross-validation. The proposed group of seven auditory features using SVM (RBF kernel) identified the concealed details with a sensitivity value of 100%, specificity value of 87.50%, and largest classification accuracy value of 92.86%, utilizing a single EEG channel (Pz). The second contribution of this research was the creation of a computer-based hidden details recognition method using single EEG channel (Pz) data. Also, the performance of the proposed approach of detecting hidden information was assessed on EEG dataset of Gao et al. (2013) related to lie recognition. The third contribution of this work was for the dataset of Gao et al. (2013), 40 features (6 time, 3 frequency, 10 wavelets, 18 EMD, and 3 correlation coefficients) in combination with SVM (RBF kernel) classification model, performed better (mean testing accuracy value of 98.80%) than the outcome (mean testing accuracy value of 95.74%) of Gao et al. (2013). The results obtained can be helpful in the recognition of concealed information in the human brain through EEG signals for practical field situations.