# Seeking arXiv endorsement for cs.CR (adversarial ML / IoT intrusion detection)

> Source: <https://discuss.huggingface.co/t/seeking-arxiv-endorsement-for-cs-cr-adversarial-ml-iot-intrusion-detection/180007#post_1>
> Published: 2026-09-07 08:13:07+00:00

I’m an MSc Computer Science student at the University of Nairobi, submitting my first paper to arXiv and need an endorsement for cs.CR.

**Paper summary:** A protocol-aware adversarial attack framework for MQTT-based intrusion detection systems. Most adversarial ML research generates feature-space attacks that ignore real network-protocol constraints, so their evasion rates don’t reflect genuine deployable threats in IoT networks. I built a three-layer constraint model (field, temporal, state-machine rules) into an evolutionary attack generator and benchmarked it against standard feature-space attacks (FGSM, PGD) on a Random Forest IDS baseline.

**Key results:** Baseline classifier: 91% accuracy, macro F1 0.86. PGD reached 24.16% evasion but 0% protocol validity (i.e., not deployable). My protocol-aware attack achieved a lower but genuine 6.00% evasion while maintaining 71.3% protocol validity (statistically significant, McNemar’s p = 8.30×10⁻¹⁷).

Happy to share the manuscript with anyone willing to take a look. If you’re arXiv-endorsement-eligible for cs.CR (or cs.LG) and interested, please reply or DM me — thank you!