Analysis and Evaluation of Keystroke Dynamics as a Feature of Contextual Authentication

The current best practice dictates that even when the correct username and password are entered, the system should look for login anomalies that might indicate malicious attempts. Most anomaly detection approaches examine static properties of user's contextual data such as IP address, screen size and browser type. Keystroke Dynamics bring additional security measure and enable us to use individuals' keystroke behaviour to decide legitimacy of the user. In this paper, we first analyze different anomaly detection approaches separately and then show accuracy improvements when we combine these solutions with various methods. Our results show that including keystroke dynamics scores in session context anomaly component as a new feature performs better than ensemble methods with different weights for session context and keystroke dynamics components. We argue that this is due to the opportunity to capture the behavioral deviations of the individuals in our augmented model.