A systematic review of federated learning based intrusion detection systems in software defined networking
Аннотация
Software Defined Networking (SDN) is susceptible to attacks that can compromise system availability. To equip a network with attack countermeasures, an Intrusion Detection System (IDS) must be integrated into the SDN architecture. Such a system collects and analyses large amounts of data, including potentially sensitive and personal data, in order to respond accurately and quickly to various types of attacks. As a result, IDS design itself raises privacy concerns. Since training is performed locally on the data sources with no external party accessing raw data, Federated Learning (FL) can reduce these risks when used to train attack and anomaly detection models. This systematic literature review, conducted following the PRISMA methodology and covering IEEE Xplore, ScienceDirect, Google Schola, Wiley, and SpringerLink, identifies 24 studies published between 2019 and 2026: 16 primary studies presenting an implemented FL-based intrusion detection system for SDN and 8 prior surveys used for comparative positioning. Of 350 records initially identified, 24 met the inclusion criteria after title/abstract and full-text screening. The systematic review shows that only 2 of the 16 primary studies implement a privacy-preserving mechanism beyond keeping training data local, indicating that privacy protection in this literature remains largely nominal rather than formally guaranteed. Reported detection accuracies reach up to 99% on balanced datasets, a condition that does not reflect the class imbalance typical of real network traffic. Additionally, this systematic review proposes a structured taxonomy for classifying existing FL-based approaches. The reviewed studies show that FL-based intrusion detection systems are a promising direction for SDN environments, while also revealing that client-scaling behavior and communication/computation overhead remain largely unmeasured, and highlighting open challenges and future research directions.