A security system for anomaly detection using probabilistic reasoning

摘要

There are currently many kinds of security software. However, there are still few systems that can detect unauthorized actions by valid account holders or by camouflage. In this paper, we propose a new security system that can identify such actions by applying a novel variation of the Naive Bayes machine learning algorithm to anomaly detection. In order to classify a user's behavior, the system uses a set of command operations of a target user to construct a profile of the users 'normal' operations. Subsequent profiles of the same or a different user's operations are then compared and classified accordingly. Results from applying the system to user distinction and anomaly detection demonstrate that it can be an effective and practical approach in real-world contexts.