Advanced Temporal-Difference Learning for Intrusion Detection∗

Nowadays intrusion detection for cyber security is the dynamically researched area. The main purpose of the intrusion detection is to distinguish normal usage of analyzed system from different forms of misuses and abnormal behaviors. The big amount of intrusion detection approaches such as soft computing and machine learning algorithms was made. In spite of visible progress, there are still many opportunities to improve state-of-the-art techniques. This paper presents the new intrusion detection technique that is based on temporal-difference learning for Markov decision processes. Actually, this method is the advanced form of the existing temporal- difference based approach named Temporal-Difference based Sequence Anomaly Detection (or TD_SAD). Due to this fact, our approach is called TD_SAD2. It is shown that the proposed approach can achieve at least comparable accuracy for intrusion detection by benchmarking with existing leading approaches.