Assessing The Robustness of Machine Learning Models for Wireless Intrusion Detection Under Adversarial Traffic Perturbations

Open Access

Notice

This is an unedited manuscript accepted for publication and provided as an Article in Press for early access at the author’s request. The article will undergo copyediting, typesetting, and galley proof review before final publication. Please be aware that errors may be identified during production that could affect the content. All legal disclaimers of the journal apply.

Year : 2026 | Volume : 04 | Issue : 01 | Page : 29 34
    By

    Siddhant Sukhatankar,

  1. Software Development Engineer-AI/ML, Demand Science Optimization Organization, Amazon, Arlington, VA, US

Abstract

As the Internet of Things (IoT) devices and wireless communication networks continue to grow rapidly, protecting systems from cyber threats has become increasingly important. Machine learning–based intrusion detection systems (IDS) have shown strong potential in detecting abnormal and malicious network activities, yet their effectiveness and resilience when facing adversarial attacks are still not sufficiently explored. This research evaluates ML models – XGBoost, Random Forest, and Multi-Layer Perceptron (MLP) – in detecting attacks within wireless IoT networks when subjected to certain specific traffic feature perturbations. Using the CICIoT2023 and IoT Intrusion datasets, we conducted binary classification experiments distinguishing benign and attack traffic. Perturbations simulating realistic adversarial manipulations were applied to numeric features at multiple levels (5%, 10%, and 25%). The results demonstrate that tree-based models, XGBoost, and Random Forest, maintain high recall under perturbation, with less than 0.2% reduction at even the highest perturbation levels, whereas MLP performance is unstable on imbalanced data. Feature importance analysis reveals that timing and protocol-related features contribute significantly to model predictions. These findings highlight the robustness of ensemble tree methods in practical IoT intrusion detection scenarios and emphasize the need for perturbation-aware evaluations for ML-based IDS. The study contributes to safer deployment of wireless IDS and provides a methodological framework for assessing model resilience against adversarial feature modifications.

Keywords: Adversarial perturbations, IoT security, machine learning, random forest, wireless intrusion detection, XGBoost

[This article belongs to International Journal of Wireless Security and Networks ]

How to cite this article:
Siddhant Sukhatankar. Assessing The Robustness of Machine Learning Models for Wireless Intrusion Detection Under Adversarial Traffic Perturbations. International Journal of Wireless Security and Networks. 2026; 04(01):29-34.
How to cite this URL:
Siddhant Sukhatankar. Assessing The Robustness of Machine Learning Models for Wireless Intrusion Detection Under Adversarial Traffic Perturbations. International Journal of Wireless Security and Networks. 2026; 04(01):29-34. Available from: https://journals.stmjournals.com/ijwsn/article=2026/view=236202


References

  1. Zhang H, Wang L, Li Q. Machine learning in IoT security: a review. IEEE Internet Things J. 2022;9(3):1234–1252.
  2. Singh R, Kaur P. DDoS attacks in IoT networks: trends and countermeasures. Comput Netw. 2023;225:109456.
  3. Chen J, Zhao Y, Xu K. ML-based intrusion detection for IoT: recent advances. J Netw Comput Appl. 2022;198:103290.
  4. Kumar A, et al. Random forest for IoT security: a comparative study. Sensors. 2023;23(4):1876.
  5. Liu S, Li W. Adversarial attacks in wireless network intrusion detection systems. IEEE Trans Inf Forensics Secur. 2022;17:789–801.
  6. Patel M, et al. Evaluating machine learning robustness for IoT traffic. Comput Commun. 2022;189:124–136.
  7. Goodfellow IJ, Shlens J, Szegedy C. Explaining and harnessing adversarial examples. In: Proc. Int. Conf. Learn. Represent. (ICLR); 2015.
  8. Canadian Institute for Cybersecurity (CIC). CICIoT2023 dataset. Fredericton (NB): University of New Brunswick; 2023.
  9. IoT intrusion dataset. Kaggle; 2022.
  10. Alotaibi A, Rassam MA. Adversarial machine learning attacks against intrusion detection systems: a survey on strategies and defense. Future Internet. 2023;15(2):62.
  11. Harbi Y, Medani K, Gherbi C, Aliouat Z, Harous S. Roadmap of adversarial machine learning in Internet of Things-enabled security systems. Sensors. 2024;24(16):5150.
  12. Adversarial attacks in IoT: a performance assessment of ML and DL models. Eng Proc. 2025;112(1):15.
  13. Towards adversarial realism and robust learning for IoT intrusion detection and classification. Ann Telecommun. 2023.
  14. Enhancing IoT security: a comparative analysis of ML and DL techniques for botnet detection. Eng Technol Appl Sci Res. 2025;15(4):24498–24505.
Regular Issue Open Access Original Research
Volume 04
Issue 01
Received 13/01/2026
Accepted 16/01/2026
Published 20/01/2026
Publication Time 7 Days
88

Login


My IP

PlumX Metrics