Using Neural Network Models for BCI Based Lie Detection

Authors:

Abstract
The focus of this paper is to investigate the performance of BCI signals as indicators for lie detection. In order to do this, six subjects’ Electroencephalography (EEG), Functional near-infrared spectroscopy (fNIRS), Heart Rate Variability (HRV), and Galvanic Skin Response (GSR) signals were collected while they were answering Boolean type questions. The EEG and fNIRS signals were gathered using g.Nautilus fNIRS-8 headset and GSR was collected by g.GSRsensor 2 , both developed by g.tec. HRV data was collected through Wellue Smart Pulse Oximeter for Adults and Infant, and the data was stored on an iPhone X using ViHealth app via Bluetooth. After processing each subjects’ BCI signals, they were stored in a CSV file and processed in MATLAB Classification Learner using the neural network models. The accuracies of the models ranged from 61.4% to 71.9%. The neural network with the 71.9% accuracy predicted 78.9% of the truths and 57.9% of the lies.