Réseaux bayésiens naïfs et arbres de décision dans les systèmes de détection d'intrusions

Bayesian networks are powerful tools for decision and reasoning under uncertainty. A very simple form of these networks is called naive Bayes, which is particularly efficient for learning and inference tasks. This paper offers an experimental study of the use of naive Bayes in intrusion detection. We show that eventhough they have a simple structure, naive Bayes provide satisfactory results. We consider three levels of attack granularities depending whether we consider whole attacks, or group them into four main categories, or just focus on normal and abnormal behaviour. Different experimentations are performed in order to find the best strategies to adopt regarding the different studied cases. We compare the performance of naive Bayes with one of well known machine learning techniques: decision trees. MOTS-CLÉS : réseaux bayésiens naïfs, arbres de décision, détection d’intrusions.