Blockchain-enabled federated learning with edge analytics for secure and efficient electronic health records management

Abstract The rapid adoption of Federated Learning (FL) in privacy-sensitive domains such as healthcare, IoT, and smart cities underscores its potential to enable collaborative machine learning without compromising data ownership. However, conventional FL frameworks face several critical challenges:...

Full description

Saved in:
Bibliographic Details
Main Authors: Munusamy S, Jothi K R
Format: Article
Language:English
Published: Nature Portfolio 2025-07-01
Series:Scientific Reports
Subjects:
Online Access:https://doi.org/10.1038/s41598-025-12225-x
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract The rapid adoption of Federated Learning (FL) in privacy-sensitive domains such as healthcare, IoT, and smart cities underscores its potential to enable collaborative machine learning without compromising data ownership. However, conventional FL frameworks face several critical challenges: high computational overhead on edge devices, significant communication latency due to frequent model updates, vulnerability to model and data poisoning attacks, and limited privacy-preserving mechanisms that expose systems to inference risks. These issues hinder the scalability, efficiency, and trustworthiness of FL in real-world, large-scale deployments—particularly in domains like Electronic Health Records (EHR) management, where data sensitivity is paramount. To address these challenges, this paper introduces the Enhanced Privacy-Preserving Blockchain-Enabled Federated Learning (EPP-BCFL) framework, which integrates blockchain with hybrid privacy mechanisms and intelligent aggregation strategies. The architecture comprises three layers: (1) an Edge Nodes Layer for on-device learning; (2) a Federated Aggregation Layer using Secure Multi-Party Computation (SMPC) and Differential Privacy (DP); and (3) a Blockchain Layer with a lightweight PoS + BFT consensus mechanism. Experimental evaluation on CIFAR-10 demonstrates 95.2% accuracy, a 43% reduction in communication latency, a 37% decrease in computational cost, and robust defense against data/model poisoning and adversarial attacks. Attack resilience improved accuracy from 72.5 to 93.2%, while privacy budget tuning achieved 90.3% accuracy at ε = 1.0. Compared to state-of-the-art models, EPP-BCFL exhibits superior performance in terms of security, scalability, and support for edge device heterogeneity, validating its applicability in secure EHR management.
ISSN:2045-2322