Work

Evaluating and Improving Network Intrusion Detection Robustness Against Adversarial SYN-Based Traffic

Year: 2026
Type: preprint
Abstract: While Network Intrusion Detection Systems (NIDS) are an integral part of the network security apparatus today, their effectiveness can diminish when malicious traffic is specifically crafted to mimic legitimate network traffic. Low-rate and threshold-evasion traffic can remain under the radar of standard traffic detection if they have features typical of denial-of-service traffic. In this study, we introduce an experimental platform that can be used to evaluate and enhance the robustness of a real-time network intrusion detection system (NIDS) against adversarial-controlled SYN traffic. The study has two parallel executions: the first is a real-time network intrusion detection system (NIDS) that performs packet-level inspection, incident reporting, rule-based detection, threat scoring, attacker profiling, automated blacklist, and forensic reporting. In the experimental validation, the system managed to process 2024 packets and produce 59 threat alerts with an average of 28.471 ms for processing one packet. The second implementation tests the detection system in controlled attack scenarios involving low, medium, and high SYN traffic and threshold-evasion attacks. For the baseline detection experiment, 75% of the attacks were detected, and one of the four attacks was not detected. An adaptive defense mechanism based on the combination of multiple features, as a function of the packet rate, number of SYN packets, ratio of SYN/ACK packets, burst behaviour, and source diversity, was then designed. All four scenarios tested were detected after the enhanced defense mechanism was implemented, and there was no missed scenario, which increased the detection rate to 100%. The results show that in a controlled experimental setting, using multiple traffic features and adaptive traffic detection methods can offer greater resilience of a network intrusion detection system (NIDS) against the evasion patterns tested in this study.
Language: English
Open Access status: green
OpenAlex ID: W7234183127
Cites:
Cited by:
Related to: 10