Machine-Learning Techniques for Detecting Attacks in SDN

TL;DR

This paper systematically benchmarks machine-learning methods for detecting malicious traffic in SDNs, highlighting limitations and proposing a foundation for more robust detection frameworks using publicly available IDS datasets.

Contribution

It provides a comprehensive analysis of existing ML techniques for SDN security and identifies key limitations to guide future research.

Findings

Classical ML methods have notable limitations in SDN security detection.

Benchmarking reveals performance gaps in current ML approaches.

Foundation laid for developing more robust SDN attack detection frameworks.

Abstract

With the advent of Software Defined Networks (SDNs), there has been a rapid advancement in the area of cloud computing. It is now scalable, cheaper, and easier to manage. However, SDNs are more prone to security vulnerabilities as compared to legacy systems. Therefore, machine-learning techniques are now deployed in the SDN infrastructure for the detection of malicious traffic. In this paper, we provide a systematic benchmarking analysis of the existing machine-learning techniques for the detection of malicious traffic in SDNs. We identify the limitations in these classical machine-learning based methods, and lay the foundation for a more robust framework. Our experiments are performed on a publicly available dataset of Intrusion Detection Systems (IDSs).