Improving Intrusion Detection System Using a New Feature Selection Technique

Intrusion detection is an important subject of research in the cyberspace field. In an Intrusion DetectionSystem (IDS), redundant and irrelevant features have a negative impact on the IDS performance. Therefore, an appropriate feature selection method is an important part of IDSs for eliminating unrelated and redundant features. In this paper, a new feature selection method is proposed that joins features level to level and step by step to select a subset of proper features in order to finally detect intrusion more accurately and speedily. The purpose of the proposed method is applying it in intrusion detection systems to distinguish a normal the connection from an intruding connection to the network. The experiments on the NSL-KDD dataset show that the proposed method in comparison with other methods selects only six important features among the 41 features in the baseline, and can detect an intrusion with precision above 99.58% by relying only on these six features. In other words, the proposed method's failure has been 42 in 10,000 connections of the network and has correctly identified other 9958 regular connections and labeled them as normal. Finally, improvement in the algorithm runtime and the percentage accuracy of the proposed method in comparison with other methods has been verified and reported.