Adaptive Clustering for Network Intrusion Detection

A major challenge in network intrusion detection is how to perform anomaly detection. In practice, the characteristics of network traffic are typically non-stationary, and can vary over time. In this paper, we present a solution to this problem by developing a time-varying modification of a standard clustering technique, which means we can automatically accommodate non-stationary traffic distributions. In addition, we demonstrate how feature weighting can improve the classification accuracy of our anomaly detection system for certain types of attacks.