A Study of the Consistency in Keystroke Dynamics

Abstract

Keystroke dynamics is the process of identifying individual users on the basis of their distinctive typing rhythms. Most current approaches implicitly assume that individual typing behavior has high consistency as well as high discriminatory ability, both of which underpin the power of the technique. However, no earlier work has been done to quantify or measure the consistency of typing behavior. This study aims to investigate the consistency of users’ typing behavior in keystroke dynamics. We obtain a keystroke benchmark dataset, propose a consistency measurement model, develop an evaluation methodology, and conduct three studies. We first quantify the consistency of users’ behavior in repeatedly typing a password, observing that a typical user’s typing behavior would become consistent over time, and changes in her typing would diminish. We then measure the consistency of keystroke timing features, finding that the combination of all features has the best consistency and smallest fluctuation. We finally examine the effect of consistency on keystroke-biometric systems, observing that the authentication performance gets better as the user’s typing behavior becomes more consistent.