SecureFed: Blockchain-Based Defence for Data Poisoning Attack in Federated Learning

Alsharidah, A; Jha, DN; Wei, B; Solaiman, E; Ranjan, R

doi:10.1109/UCC63386.2024.00064

SecureFed: Blockchain-Based Defence for Data Poisoning Attack in Federated Learning

Lookup NU author(s): Ahmad Alsharidah, Dr Dev Jha ORCiD, Dr Bo Wei ORCiD, Dr Ellis Solaiman ORCiD, Professor Raj Ranjan

Downloads

Accepted version [.pdf]

Licence

This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

Abstract

Federated Learning (FL) is a distributed machine learning paradigm that enables collaborative model training across multiple devices without sharing raw data. In Industrial Internet of Things (IIoT) environments, where devices are heterogeneous and distributed, FL can enhance privacy and reduce communication overhead. However, the decentralized nature of FL makes it vulnerable to data poisoning attacks, such as label-flipping attacks (LFA), which can compromise the integrity of the global model. This paper proposes SecureFed, a Blockchain-based FL framework that employs hybrid defence techniques to counter LFA. Using Cosine similarity for initial anomaly detection, suspicious updates are moved to a validation layer for further investigation. The blockchain layer ensures secure and transparent verification of updates, addressing the scalability and computational limitations of traditional methods. SecureFed is evaluated using the Fashion-MNIST dataset, running multiple experiments. The results show that SecureFed improves the overall model accuracy by around 10% compared to classic FL, demonstrating its robustness against malicious attack handling.