BehavioralAuthentication

We present an approach to a mostly unsupervised user authentication and identification based on mouse dynamics. Our hypothesis is that one can successfully identify a user on the basis of cursor movements. Our system identifies a user as unauthorized if the behavior within a 10 seconds period deviates sufficiently from the learned behavior of an authorized user. Our results for four users show that we can identify these as unauthorized based on their cursor dynamics with a false positive rate of 0% and a false negative rate of 20% on the authorized user data. Nevertheless, we have to research more thoroughly and use more data to validate our results. We point out that analysing cursor dynamics alone is not yet sufficient for a practical user authentication system.