Performance Analysis of Keystroke Dynamics Using Classification Algorithms

Abstract

Authentication is the process of verifying the identity of a user. Biometric authentication assures user identity by identifying users physiological or behavioral traits. Keystroke dynamics is a behavioral biometric based on users typing pattern. It can be used to authenticate legitimate users based on their unique typing style on the keyboard. From a pattern recognition point of view, user authentication using keystroke dynamics is a challenging task. It can be accomplished by using classification algorithms - two-class and one-class classification algorithms. In this paper, we study and evaluate the effectiveness of using the one-class classification algorithms over the two-class classification algorithms for keystroke dynamics authentication system. We implemented and evaluated 18 classification algorithms (both two-class and one-class) from the literature of keystroke dynamics and pattern recognition. The result of our experiments is evaluated using 28 subjects with the total of 378 unique comparisons for each classifier. Our results show that the top-performing classifiers of one-class are not very different from two-class classifiers and can be considered to use in the real-world authentication systems.