A Comparative Analysis of Machine Learning Algorithms in Detecting Deceptive Behaviour in Humans using Thermal Images

Abstract

Various researches are going on to identify the deceptive behaviour in humans by means of modern technology. The traditional and long term method used for detection is the polygraph technology, however it is not considered reliable. Impressive research works have been made in the last few decades to detect the deceptional behaviour in humans in a most promising way. Thermal imaging technology which is a rising trend can be applied to figure out the stress levels in humans by assessing the heat radiated from facial region. In this approach we record the difference in temperature in the pre-orbital region to discriminate between a truth teller and a liar. This paper aims at a comparative study of machine learning algorithms on a facial thermal image data set. By using the three algorithms K-Nearest Neighbor, Logistic Regression and Decision Tree we attempted to create a method that can detect lies. We arrived at a solution that gave 90% accuracy for K-Nearest Neighbor, Logistic Regression each and 85% accuracy for Decision Tree.