Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

SFA-IoT: segment-to-flow aligned adversarial defense for interpretable IoT device identification.

Sayed Jahed Hussini1, Raeed Al-Sabri2, Ala Al-Fuqaha2

  • 1Department of Computer Science, Western Michigan University, Kalamazoo, MI, USA. sayedjahed.hussini@wmich.edu.

Scientific Reports
|June 26, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

FEGGNN: Feature-Enhanced Gated Graph Neural Network for robust few-shot skin disease classification.

Computers in biology and medicine·2025
Same author

MGATAF: multi-channel graph attention network with adaptive fusion for cancer-drug response prediction.

BMC bioinformatics·2025
Same author

IoT based intelligent pest management system for precision agriculture.

Scientific reports·2024
Same author

AutoAMS: Automated attention-based multi-modal graph learning architecture search.

Neural networks : the official journal of the International Neural Network Society·2024
Same author

AutoDDI: Drug-Drug Interaction Prediction With Automated Graph Neural Network.

IEEE journal of biomedical and health informatics·2024
Same author

Deep learning based multimodal urban air quality prediction and traffic analytics.

Scientific reports·2023

We introduce SFA-IoT, a novel adversarial purification framework for Internet of Things (IoT) device identification. This method enhances model robustness against adversarial attacks by leveraging unique properties of IoT traffic, significantly restoring identification accuracy.

Area of Science:

  • Cybersecurity
  • Machine Learning
  • Internet of Things (IoT)

Background:

  • Machine learning (ML) models for IoT device identification excel on clean data but are vulnerable to adversarial perturbations.
  • Existing models lack transparency, offering limited insight into failure mechanisms under attack.
  • Adversarial attacks pose a significant threat to the reliability of IoT device identification systems.

Purpose of the Study:

  • To propose SFA-IoT, an adversarial purification framework to enhance the robustness of ML-based IoT device identification.
  • To address the performance degradation of IoT identification models when subjected to adversarial perturbations.
  • To provide insights into the failure modes of these models and guide defense strategies.

Main Methods:

  • Leveraging an observed property of IoT traffic where adversarial perturbations concentrate in specific protocol-event regions.
Keywords:
Adversarial machine learningDevice identificationExplainable artificial intelligenceInternet of ThingsRobustnessSegment-to-flow alignment

Related Experiment Videos

  • Implementing SFA-IoT with segment-to-flow alignment using a frozen expert identifier and dual identity constraints (feature and embedding levels).
  • Evaluating four classifiers (DNN, SVM, XGBoost, Decision Tree) against white-box (FGSM, PGD) and black-box attacks on the UNSW IoT dataset.

    • Main Results:

    • FGSM and PGD attacks reduced classifier accuracy by up to 92% on the UNSW IoT dataset.
    • SFA-IoT successfully restored 9-40% of accuracy in a single forward pass, with a peak improvement of 58.6% for decision trees.
    • SHAP and LIME diagnostics were used for explanation and to inform the defense architecture design.

    Conclusions:

    • SFA-IoT significantly improves adversarial robustness for IoT device identification models.
    • The framework effectively mitigates performance degradation caused by adversarial perturbations.
    • Interpretability methods can guide the development of more robust and explainable AI systems, demonstrating that interpretability and adversarial robustness are complementary objectives.