Clustering di-graphs for continuously verifying users according to their typing patterns

Traditionally users are authenticated based on a username and password. However, a logged station is still vulnerable to imposters when the user leaves her computer without logging off. Keystroke dynamics methods can be useful to continuously verify a user, after the authentication process has successfully ended. Within the last decade several studies proposed the use of keystroke dynamics as a behavioral biometric tool to verify users. We propose a new method, for compactly representing the keystroke patterns by joining similar pairs of consecutive keystrokes. The proposed method considers clustering di-graphs based on their temporal features. The proposed method was evaluated on 10 legitimate users and 15 imposters. Encouraging results suggest that the proposed method detection performance is better than that of existing methods. Specifically we reach a False Acceptance Rate (FAR) of 0.41% and a False Rejection Rate (FRR) of 0.63%.