Federated Learning Framework Based on Distributed Storage and Diffusion Model for Intrusion Detection on IoT Networks
The integration of Internet of Things (IoT) devices into smart environments has become increasingly prevalent, resulting in the collection of valuable user and service data. However, effectively utilizing this data often requires its aggregation on a central server to train algorithms capable of ide...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10979891/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The integration of Internet of Things (IoT) devices into smart environments has become increasingly prevalent, resulting in the collection of valuable user and service data. However, effectively utilizing this data often requires its aggregation on a central server to train algorithms capable of identifying and preventing malicious attacks, such as reconnaissance, DoS (Denial of service), DDoS (Distributed denial of service) within IoT networks. This transmission of raw data not only incurs substantial bandwidth costs but also raises significant privacy concerns. In this paper, we propose a federated learning framework for intrusion detection on IoT networks that incorporates a distributed storage system based on the Ethereum blockchain, enhancing the security of the federated learning process. This design offers several key benefits, including scalability, high availability, redundancy, and the capacity to process large datasets. Despite these advantages, relying solely on federated learning may not yield accurate results, particularly when dealing with highly imbalanced datasets. To address this challenge, we have integrated a diffusion model for data augmentation at each local node, which strengthens model robustness. Furthermore, to protect data privacy at each local node, we utilize transmitting and averaging model parameters instead of raw data. The proposed framework is trained and evaluated in two datasets. The MNIST (Modified National Institute of Standards and Technology) dataset and BoT-IoT dataset. Our results indicate significant improvements in detecting zero-day attacks, achieving an average F1-score of 98.3% on the short version of the BoT-IoT dataset as well. |
|---|---|
| ISSN: | 2169-3536 |