Improving Anomaly Detection Error Rate by Collective Trust Modeling

Current Network Behavior Analysis (NBA) techniques are based on anomaly detection principles and therefore subject to high error rates. We propose a mechanism that deploys trust modeling, a technique for cooperator modeling from the multi-agent research, to improve the quality of NBA results. Our system is designed as a set of agents, each of them based on an existing anomaly detection algorithm coupled with a trust model based on the same traffic representation. These agents minimize the error rate by unsupervised, multi-layer integration of traffic classification. The system has been evaluated on real traffic in Czech academic networks.