P300 Based Deception Detection using Convolutional Neural Network

Algorithms for automatic lie detection from analysis of brain signals have remained a question of interest that fascinate the research community. Nevertheless, the detection algorithms still lack robustness while processing brain signals. The aim would be to learn whether a person is deceitful or not by detection techniques and suggest a vigorous lie detection algorithm. Moreover, recently proposed algorithms for lie detection have shown to achieve a classification accuracy of around 96%. While different classification algorithms such as Support Vector Machines, multilaver neural network, Extreme Learning Machine, and Linear Discriminant Analysis have proposed which typically utilizes three different types of features like time domain features, frequency domain features, and wavelet features, are anticipated in the literature. Accordingly, in this research paper, we presented a lie detection system front the P300 wave. For automatic optimum feature learning, we applied an approach from deep learning, Convolutional Neural Network (CNN), which is very effective for classification problems. The presented model significantly achieves high accuracy of 99.6%. The experimental outcomes show that the technique put forward achieves the maximum accuracy with a lesser amount of training and testing time and reveal improved performance. Additionally, an all-inclusive discussion on the choice of appropriate CNN architecture and classification results presented in this paper along with a comparison with the prior approaches of lie detection.