Pattern construction by extracting user specific features in keystroke authentication system

One of the biggest challenges in behavioral biometric systems is to select biometric features for pattern construction due to large domain of user's behavior. It is desirable that, behavioral pattern should be as small as possible for low computational cost of the system. But reduced pattern size costs accuracy of the authentication system. In this paper, authors have proposed to select user specific features for pattern construction, which is an obvious choice, as specific features are much stronger than others. This paper also proposes a technique to select user specific feature set for keystroke dynamics system. It is also demonstrated through experiments that, when user specific features are used, pattern size is drastically reduced without compromising authentication accuracy of the system.