An Adaptive Hybrid Architecture for Intrusion Detection Based on Fuzzy Clustering and RBF Neural Networks

Automatic detection of network intrusion is a challenging task because of increasing types of attacks. Many of the existing approaches either are rigid, inflexible designs tailored to a specific situation or require manual setting of design parameters such as the initial number of clusters. In this paper we allow the design parameters to be determined dynamically by adopting a layered hybrid architecture, hence resolving the aforementioned shortcomings. The first layer uses FCM and GK fuzzy clustering to extract the features and the second layer uses a set of RBF neural networks to perform the classification. The flexible design parameters are initial number of clusters, number of RBF networks and number of neurons inside each network which are determined with minimal input from the user. The simulation result shows high detection rates as well as fewer false positives compared to earlier approaches.