Intrusion Detection Systems (IDSs) have the reputation of generating many false positives. Recent approaches, known as stateful IDSs, take the state of communication sessions into account to address this issue. A substantial reduction of false positives, however, requires some correlation between the state of the session, known vulnerabilities, and the gathering of more network context information by the IDS than what is currently done (e.g., configuration of a node, its operating system, running applications). In this paper we present an IDS approach that attempts to decrease the number of false positives by collecting more network context and combining this information with known vulnerabilities. The approach is model-driven as it relies on the modeling of packet and network information as UML class diagrams, and the definition of intrusion detection rules as OCL expressions constraining these diagrams. The approach is evaluated using real attacks on real systems, and appears to be promising.

, ,
Department of Systems and Computer Engineering

Massicotte, F. (Frederic), Couture, M. (Mathieu), Briand, L. (Lionel), & Labiche, Y. (2007). Model-driven, network-context sensitive intrusion detection.