EUVIMEDEUVIMED entdecken
Crossref

Privacy-Preserving Network Traffic Generation and Domain Transfer for Data-Scarce IoT Environments

Minjae Kang, Gunhee Cho, Sungju Yun, Nahyun Kim, Jungmin Lee, Yeonjoon Lee

IEEE Internet of Things Journal · 2026-11 · 13 (22) · 53707-53723

DOI: 10.1109/jiot.2026.3730008 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Internet of Things Journal, 2026-11. Verfasst von Minjae Kang und 5 weiteren Autor:innen. Die Arbeit nennt 82 Literaturangaben. Gefördert von: Korea Research Institute for Defense Technology Planning Advancement (KRIT) grant funded by Korea Government through the Defense Acquisition Program Administra….

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Artificial intelligence (AI)-powered security systems in Internet of Things (IoT) environments require high-quality network traffic data for effective training. However, data collection faces barriers, including high costs, restricted access to isolated network environments, and privacy regulations. As a result, training data is limited in quantity and diversity. To address data scarcity, researchers have explored synthetic generation methods to create realistic network traffic. However, existing methods suffer from three challenges. First, they fail to capture temporal dynamics of network traffic, which comprises mixed-type features requiring specialized preprocessing. Second, synthesizing high-fidelity traffic requires substantial training data, which is scarce in specialized IoT environments. Third, they lack formal privacy mechanisms to protect sensitive network information. To address these challenges, we propose NetTransfer, a framework with three targeted solutions. For the first challenge, we design feature-unified preprocessing to process mixed-type features into consistent representations, so the diffusion model can effectively capture temporal dynamics. For the second challenge, we propose dynamic adaptive network transfer (DANT) that enables efficient domain transfer with limited data. To address the third challenge, we develop selective-differential privacy (DP) that provides sample-level DP for fixed-length temporal traffic windows, rather than packet- or flow-level privacy, with reduced performance impact. Evaluations across six network datasets show that NetTransfer reduces distributional distance by up to 70% compared with state-of-the-art baselines while preserving protocol-field validity and downstream classification utility. In domain transfer experiments, NetTransfer adapts effectively with limited target data across multiple target domains. Privacy analysis shows sample-level protection with less than 5% fidelity degradation.

Förderung

  • Korea Research Institute for Defense Technology Planning Advancement (KRIT) grant funded by Korea Government through the Defense Acquisition Program Administra…

Zitieren

APA

Kang, M., Cho, G., Yun, S., Kim, N., Lee, J., & Lee, Y. (2026). Privacy-Preserving Network Traffic Generation and Domain Transfer for Data-Scarce IoT Environments. IEEE Internet of Things Journal, 13(22), 53707-53723. https://doi.org/10.1109/jiot.2026.3730008

Vancouver

Kang M, Cho G, Yun S, Kim N, Lee J, Lee Y. Privacy-Preserving Network Traffic Generation and Domain Transfer for Data-Scarce IoT Environments. IEEE Internet of Things Journal. 2026;13(22):53707-53723. doi:10.1109/jiot.2026.3730008

BibTeX
@article{Kang2026,
  author = {Kang, Minjae and Cho, Gunhee and Yun, Sungju and Kim, Nahyun and Lee, Jungmin and Lee, Yeonjoon},
  title = {Privacy-Preserving Network Traffic Generation and Domain Transfer for Data-Scarce IoT Environments},
  journal = {IEEE Internet of Things Journal},
  year = {2026},
  volume = {13},
  number = {22},
  pages = {53707--53723},
  doi = {10.1109/jiot.2026.3730008}
}
Kostenlos starten →