A Review on Federated Learning-Based Network Intrusion Detection System

Abstract

The security of network-connected devices is a subject of increasing concern due to the widespread use of the Internet in recent years and the possibility of numerous flaws that may be exploited by an attacker. Mobile phones, wearable technology, and self-driving cars are just a few examples of distributed networks that produce and transmit enormous amounts of data daily. The security and privacy of such devices are significantly enhanced by intrusion detection systems. For well-known intrusion detection systems (IDS) to detect increasingly sophisticated cybersecurity assaults effectively and efficiently, machine learning (ML) techniques must be applied. Due to their success in achieving high classification accuracy, these techniques have shown themselves to be quite useful. Thus, many solutions were developed to provide protection against cyberattacks and intruders. Many papers from many reputed authors were studied to understand the working of these solutions. However, the requirement to store and transmit data to a centralized server may put privacy and security concerns in jeopardy. Federated learning (FL), a privacy-preserving decentralized learning strategy that trains models locally and sends the parameters to a centralized server, fits in well to reduce privacy concerns associated with centralized systems. A computational methodology for networked machine learning called federated learning enables numerous cooperating organizations to train a single large-scale model. The rest of this review paper goes in-depth about the various solutions proposed by multiple authors. Their methodology, results, advantages, and disadvantages are analyzed, compared, and contrasted.