Keystroke dynamics based authentication system with unrestricted data collection

Abstract

In today's world, superior authentication mechanisms have gained utmost importance to deal with increased synthetic forgeries. One such authentication mechanism uses keystroke dynamics to uniquely label users on the basis of their typing pattern. This paper proposes a statistical approach to implement such a keystroke dynamics based authentication system. The primary focus in this paper while building an authentication system is on providing an unrestricted environment for data collection. However, removing restrictions makes the data unfit for direct feature extraction, and therefore, it requires preprocessing. The majority of this paper presents techniques for improved data preprocessing with its principal element being the removal of outlier values from the dataset. It presents typing trajectories of a particular user before and after removing outliers from the dataset, to show that the similarities in typing pattern become even more prominent after preprocessing. After using the proposed method for removing outliers, the authentication accuracy of the system is increased many folds.