EFIS: Evolvable-neural-based fuzzy inference system and its application for adaptive network anomaly detection

Abstract

This paper presents an application of a new type of fuzzy inference system, denoted as evolvable-neural-based fuzzy inference system (EFIS), for adaptive network anomaly detection in the presence of a concept drift problem. This problem cannot be avoided to happen in every network. It is a problem of modeling the behavior of normal traffic while it keeps changing over time in continuous manner. EFIS can solve the concept drift problem by having dynamic network traffic profile creation and adaptation. The profile is then being further used to detect anomaly. An enhanced evolving clustering method (ECMm), which is employed by EFIS for online network traffic clustering, is also presented. It is demonstrated, through experiments, that EFIS can evolve in a growing network and also successfully detect network traffic anomalies. © Springer-Verlag Berlin Heidelberg 2006.