A1 Refereed original research article in a scientific journal
FBMP-IDS: FL-based Blockchain-powered Lightweight MPC-secured IDS for 6G networks
Authors: Sakraoui, Sabrina; Ahmim, Ahmed; Derdour, Makhlouf; Ahmim, Marwa; Namane, Sarra; Dhaou, Imed Ben
Publisher: IEEE
Publication year: 2024
Journal: IEEE Access
Journal name in source: IEEE Access
Volume: 12
First page : 105887
Last page: 105905
eISSN: 2169-3536
DOI: https://doi.org/10.1109/ACCESS.2024.3435920
Web address : https://ieeexplore.ieee.org/document/10614597
Self-archived copy’s web address: https://research.utu.fi/converis/portal/detail/Publication/457406803
The coming 6G wireless network is poised to achieve unprecedented data rates, latency, and integration with newer technologies like AI and IoE. On the other hand, along with this kind of growth in the AI domain and the large-scale connectivity in 6G, it is also going to raise many security concerns at the level of intrusion detection and prevention. For intrusion detection, centralized approaches won’t be able to work effectively, therefore there is an utmost need to design decentralized and privacy-preserving solutions. In this work, we propose a novel secure gradients exchange algorithm for distributed intrusion detection in 6G networks. Our method is designed, to take into account the use of Federated Learning with secure multi-party computation and blockchain technology to ensure that collaborating parties are able to conduct training of intrusion detection models in a secure and collaborative way by retaining privacy in the data. Gradient compression and adaptive secure aggregation strategies are used to further optimize communication overhead and computational complexity so that our design works in a robust and efficient manner with the high data rates and huge connectivity that 6G networks will provide. To achieve our goal, experiments using the CICIoT2023 dataset were performed, and results showed that a federated learning-based hybrid model composed of CNN1D and a multi-head attention mechanism outperformed other well-known deep learning models in terms of performance. It achieved the highest average accuracy with 79.92%, the highest average detection rate with 77.41%, and a low false alarm rate with 2.55%.
Downloadable publication This is an electronic reprint of the original article. |  |
