A comprehensive survey of machine learning-based network intrusion detection

Abstract

In this paper, we survey the published work on machine learning-based network intrusion detection systems covering recent state-of-the-art techniques. We address the problems of conventional datasets and present a detailed comparison of modern network intrusion datasets (UNSW-NB15, TUIDS, and NSLKDD). Recent feature-level processing techniques are elaborated followed by a discussion on supervised multi-class machine learning classifiers. Finally, open challenges are pointed out and research directions are provided to promote further research in this area.