Agent Based Intrusion Detection with Soft Evidence

摘要

In this paper we propose a new framework for intrusion detection, called Probabilistic Agent-Based Intrusion Detection (PAID), using agent encapsulated Bayesian networks. It allows agents to share their beliefs, i.e., the calculated probability distribution of event occurrence. A unique feature of our model is that the agents use the soft evidential update method to process beliefs. This provides a continuous scale for intrusion detection, supports merging of signature based and anomaly based systems, and reduces the communication overhead in a distributed intrusion detection scenario. We have developed a FIPA compliant agent communication architecture that provides a prototype implementation.