Skip to content
Open access

Exposing and Exploiting the Ineffectiveness of Real-Time ML-Based NIDSs on Ambiguous Traffic in Industrial IoT Networks

2026 · Computers, Materials & Continua · 0 citations · 24 references

Abstract

: Pure TCP ACK packets, defined as acknowledgments without payload, are inherently ambiguous in network traffic classification. Their header fields are dictated entirely by TCP connection state, rather than application behavior, so field value variations do not contain behavior semantics that would reliably distinguish benign from malicious traffic. Such packets are prevalent in Industrial Internet of Things (IIoT) environments and occur at high volumes due to the communication characteristics of domain-specific protocols. These properties motivate an investigation into the effectiveness of Machine Learning Network Intrusion Detection Systems (ML-NIDSs) in IIoT networks for continuous monitoring of network activity. Our empirical evaluation demonstrates that state-of-the-art real-time detection models consistently misclassify pure ACK packets, across both benign and attack classes, with high confidence. To illustrate the security implications, we introduce a novel attack variant, the DUP ACK Flooding attack , which deliberately exploits this blind spot. We demonstrate its practical feasibility through testbed experiments under real-world conditions and analyze its impact on security monitoring and network operations. Finally, an ablation study in which pure ACK packets are withheld during training and introduced only at inference shows that the approach does not fully mitigate the issue, highlighting the need for further research.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.