Using machine learning for lie detection: classification of human visual morphology

Abstract

Though there exists different methods of detecting lies, machine learning can be used to create a reliable and more efficient system to detect lies. This thesis proposes a method of using nonverbal human behaviors to detect lies using machine learning. This lie detection system is based on micro-expressions of human beings which uses Facial Landmark Detection System and Azure Machine Learning. Movements of individual facial muscles are recorded while a person answers some certain questions. By using the two algorithms Two-Class Support Vector Machine and Linear Regression, we attempted to create a machine that can detect lies. We reached an accuracy of approximately 76.2%.