Securing FL-Enabled NWDAF in Private 5G Networks Against Hybrid Attacks

Xiong, Ruoting ORCID: https://orcid.org/0000-0003-0575-4218, Xiong, Maolin, Ren, Yi ORCID: https://orcid.org/0000-0001-7423-6719, Ren, Wei and Parr, Gerard ORCID: https://orcid.org/0000-0002-9365-9132 (2026) Securing FL-Enabled NWDAF in Private 5G Networks Against Hybrid Attacks. IEEE Transactions on Green Communications and Networking, 10. pp. 4392-4407.

[thumbnail of FL__IEEE_Trans__on_Green (7)]
Preview
PDF (FL__IEEE_Trans__on_Green (7)) - Accepted Version
Available under License Creative Commons Attribution.

Download (1MB) | Preview

Abstract

5G private mobile networks (PMNs) are gaining popularity, however, the resulting data silos can limit the effectiveness of network intelligence. 3GPP has introduced federated learning (FL)-based network data analytics function (NWDAF) to mitigate data silos in 5G PMNs. However, this architecture remains vulnerable to both inference and poisoning attacks, which undermine its efficiency and utility. Existing solutions incur high computational overhead and limited scalability, making them less suitable for resource-constrained NWDAF environments. In this paper, we implemented the FL-enabled NWDAF architecture, where multiple NWDAFs in different 5G PMNs train a model cooperatively for Apps identification (e.g., further for flow control). Besides, we designed a matrix-based ciphertext distance calculation (MCDC) approach combined with Krum model selection method for solving both poisoning attacks and inference attacks in FL framework. Experimental results show that the MCDC method is well suited to FL settings, maintaining model accuracy while applying protection. The models achieve 95-98% accuracy on homogeneous dataset, and 85-92% accuracy on heterogeneous dataset. Furthermore, the proposed scheme achieves lower computational latency and memory overhead compared with existing state-of-the-art methods. When combined with the Krum algorithm, the proposed framework maintains robust model aggregation even with 30% malicious clients, as low-quality updates can be effectively filtered out.

Item Type: Article
Uncontrolled Keywords: 5g private network,federated learning,inference attacks,network data analytics function,poisoning attacks,renewable energy, sustainability and the environment,computer networks and communications,das not required, no data associated with output ,/dk/atira/pure/subjectarea/asjc/2100/2105
Faculty \ School: Faculty of Science > School of Computing Sciences
Faculty of Science
UEA Research Groups: Faculty of Science > Research Groups > Cyber Intelligence and Networks
Faculty of Science > Research Groups > Data Science and AI
Faculty of Science > Research Groups > Health Computing
Related URLs:
Depositing User: LivePure Connector
Date Deposited: 19 Nov 2026 10:09
Last Modified: 19 Nov 2026 10:09
URI: https://ueaeprints.uea.ac.uk/id/eprint/105148
DOI: 10.1109/TGCN.2026.3735240

Downloads

Downloads per month over past year

Actions (login required)

View Item View Item