Collaborative approach to network behaviour analysis based on hardware-accelerated FlowMon probes

Network behaviour analysis techniques are designed to detect intrusions and other undesirable behaviour in computer networks by analysing the traffic statistics. We present an efficient framework for integration of anomaly detection algorithms working on the identical input data. This framework is based on high-speed network traffic acquisition subsystem and on trust modelling, a well-established set of techniques from the multi-agent system field. Trust-based integration of algorithms results in classification with lower error rate, especially in terms of false positives. The presented system is suitable for both online and offline processing, and introduces a relatively low computational overhead compared to deployment of isolated anomaly detection algorithms.