Biometrics in a Data Flow Context with Immune Algorithms

Abstract

The growing presence of the Internet day-to-day tasks, together with the evolution of the computing systems, led to a greater data exposure. This scenario highlights the need for safer user authentication systems. A promising approach for user authentication is the use of biometric technologies, which analysis user's physiological and behavioral features. Biometric technologies have less vulnerabilities than passwords or cards for authentication, which may be stolen or even cloned. An important feature of biometric data is that they may undergo small changes through time for the same user. Due to that, classifiers which adopt a static approach may reduce their predictive performance through time, as they would not adapt to these changes. In machine learning, an area which involves related concepts is data flow mining. The mentioned phenomena is known as concept drift in the area of data flow mining. It is possible to draw a parallel between the user profile modeling with biometrics and the functioning of the artificial immune systems, a subarea of computational intelligence widely used in machine learning. Both need to identify what is normal in order to find deviations, which would be potential attacks. This parallel shows that the application of immune algorithms is a promising alternative for the user recognition by biometric means. In fact, these algorithms attained good performance in previous works in the area. However, an aspect which has been little explored is the study of these algorithms in a scenario which concept drift occurs, as mentioned. The proposal of this work is to combine concepts of data flow mining with immune algorithms for the user profile modeling with biometrics, considering the fact that concept drift may arise.