A Review on Robustness in Machine Learning Based Network Intrusion Detection Systems

Abstract

Modern Network Intrusion Detection Systems require constant adaptation to new attacks, making the application of Machine Learning (ML) based solutions attractive. The application of such systems in real-world applications require robust solutions that work well under data alterations and are robust against adversarial attacks. The literature includes several works focusing on the robustness of (ML-based) Network Intrusion Detection System (NIDS) models, but it remains unclear which aspects of robustness are relevant, well-researched, or need of further investigation. Furthermore, the existing literature lacks a structured overview due to the varying experimental setups, particularly the differences in the benchmark datasets, NIDS models, and the scope of robustness.
This work aims to close this gap by reviewing recent studies on the robustness of ML-based intrusion detection from 2021 to 2025, with the goal of structuring the landscape of experimental setups and the varying definitions and scopes of robustness. Our review reveals significantly different setups that hinder result comparison, making it difficult for practitioners to select appropriate approaches and for researchers to establish consistent benchmarks and advance development. To address this, we propose benchmarking setups for robustness in ML-based NIDS systems to enable a systematic comparison of approaches.


Mehr zum Titel

Titel A Review on Robustness in Machine Learning Based Network Intrusion Detection Systems
Medien Computer Networks
Verlag Elsevier
Heft 287
Verfasser Maximilian Wolf, Prof. Dr. Dieter Landes, Andreas Hotho, Daniel Schlör
Seiten 112467
Veröffentlichungsdatum 01.09.2026
Projekttitel BRACE-LLM
Zitation Wolf, Maximilian; Landes, Dieter; Hotho, Andreas; Schlör, Daniel (2026): A Review on Robustness in Machine Learning Based Network Intrusion Detection Systems. Computer Networks (287), 112467.