Deception Detection Using a Multichannel Custom-Design EEG System and Multiple Variants of Neural Network

Abstract

This study aims to develop a BCI system using Electroencephalogram (EEG) for lie detection. EEG signal is obtained in a non-invasive manner with eight electrodes mounted at designated positions on the scalp employing a hand-made design EEG headset. The selected input for classification in this research comprises 11 features extracted across all three domains: time, frequency, and time-frequency. In which nine time-domain features include: Mean, Variance, Standard Deviation, Skewness, Kurtosis, Permutation Entropy, SVD Entropy, Approximate Entropy, and Sample Entropy. Over the frequency domain and time-frequency domain, we have the Spectral Entropy and an 8-dimensional image computed by Continuous Wavelet Transform. Modern researches indicate that Neural Networks have an extraordinary capacity to classify patterns with diverse inputs such as data sequences and pictures. Four variations are applied in this study, including the Multilayer Perceptron (MLP), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network (CNN). In which MLP, LSTM, GRU are worked on the first ten time and frequency features, while the 8-dimensional image composed after the Continuous Wavelet Transform is adopted as the input of the CNN network. The evaluation results demonstrated the feasibility of our EEG-Based BCI system in Deception recognition.

This work was supported by an NRF grant of Korea Government (2019R1A2C1089139).