Specific attack adjusted Bayesian network for intrusion detection system
Milan Tuba, Dušan Bulatović, Olga Miljković, Dana Simian · 2008
This paper further examines suitability of Bayesian networks for intrusion detection in computer networks. It is a continuation of [1]. Medical records should be readily available but also well protected, which are contrdictory goals. Automatic intrusion detection system (IDS) is required, but all used approaches have certain shortcomings. Bayesian networks are known to have good features, except that in general case are computationally too expensive. Here we show that limiting intrusion detection to any specific attack makes Bayesian networks manageable and suitable for the IDS. Example network is constructed and examined. Results are promising since with very limited computation and low sensitivity to the quality of prior knowledge, potentially dangerous situations are successfully detected and classified. Such Bayesian network can represent an independent agent in a distributed system.