Machine and Deep Learning Applications to Mouse Dynamics for Continuous User Authentication

Abstract

Static authentication methods, like passwords and PINs, authenticate a user once and only once. Continually authenticating a user even after initial access can drastically increase account security against imposters. Unique behaviors, such as mouse movements, are distinct and varied enough between humans to be irreproducible. This makes them a viable biometric to utilize for user authentication. Machine and deep learning have exploded in popularity due to their superior ability to process large amounts of data. We train and evaluate three machine learning and three deep learning algorithms on our own novel mouse dynamics dataset.