A Review on Machine Learning based Security Approaches in Intrusion Detection System

Abstract

The rapid growth and advances in communication and Internet technologies result in efficient business transactions, communications as well as collaborations. However, this global system of interconnected networks increases the visibility and transparency of resources and data. As a result, threats actors utilize the vulnerabilities of networks to perform malicious activities and pose challenges for security. An Intrusion Detection System (IDS) is a software/tool to detect intrusions in network to protect resources by inspecting the traffic. Threat actors evolve with new techniques and technologies to release new threats for information technology resources. Threats are dynamic in nature and type. A traditional signature and experts rules-based techniques are no longer sufficient to detect ever-evolving threats. Machine Learning (ML) algorithms play a vital role in network cyber security. The two classifications of ML algorithms are shallow modes and deep learning (DL) models. These techniques increase the accuracy rate and reduce false rates on intrusion detection. ML techniques boost the performance of the IDS significantly. In this work, we perform a literature review of existing studies on ML-based IDSs. As a result, we present our findings and potential future research directions in this work. From the literature review, we revealed deep learning models outperform shallow ML models according to the evaluation measures such as execution time, complexity, accuracy and error rate. Moreover, we highlighted various research challenges in intrusion detection and also proposed potential solutions to resolve those challenges.

Topics & Concepts