Adaptive Multiagent System for Network Traffic Monitoring

Abstract

An application of agent-based data mining for near-real time detection of attacks against the computer networks and connected hosts is based on processing network traffic statistics provided by high-speed network monitoring cards and using a set of known anomaly-detection techniques to identify the anomalous behavior. The individual anomaly-detection methods have relatively high error rates that make them unfit for most practical deployments. Using the agent-based trust modeling technique, the Camnep system fuses the data provided by anomaly-detection methods and progressively builds a better classification with an acceptable error rate. The system uses agent-based self-adaptation techniques to dynamically align its structure with the changes in network traffic structure and attacks.