Privacy-preserving Blockchain-enabled Federated Learning with Hierarchical Clustering for Electronic Health Records

Read full text Summarise with VIRAT
Authors: Shweta Madiwalar, Jayashree C. Nidagundi
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The speedy implementation of Federated Learning (FL) in privacy-sensitive areas such as healthcare, IoT, and smart cities highlights its ability to facilitate shared AI-based modelling while preserving ownership.Nevertheless, Existing frameworks are confronted with various issues like computational overhead, prolonged communication latency caused by iterative parameter exchanges, and exposure to adversarial threats such as data or model tampering.Furthermore, the absence of robust privacy preserving mechanisms amplifies the risk of information leakage.Collectively, these constraints obstruct the scalability, performance efficiency, and dependability of federated learning in large-scale, mission critical environments, most notably in domains like Electronic Health Record (EHR) management.To address these issues, this research work proposes EPP-CHBCFL (Enhanced Privacy-Preserving Clustered Hierarchical Blockchain-Enabled Federated Learning) framework.The proposed framework combines Blockchain technology with hybrid privacy strategies and dual-layer aggregation methods.The proposed architecture consists of tri-layer structure: (1) a local training layer for on-device learning (2) a Federated Aggregation Layer that combines clustered and hierarchical methods with SMPC and DP for secure federation and (3) a blockchain layer that employs PoS and BFT consensus to validate updates securely.This work modifies FL by adding a clustered hierarchical step to group customers based on the local updates with the global model .After clustering, training is conducted separately for each group concurrently .We demonstrate that the proposed model leads to faster model convergence with reduced communication overhead especially in the context of non-iid condition in contrast to baseline FL without clustering.Furthermore, the frame work enables a greater proportion of clients to reach the desired accuracy relative to conventional FL.Experimental evaluation on the MIMIC-III clinical dataset show accuracy of 98.6%, 38% reduction in communication latency, a 30% decrease in computational costs, and strong defenses against data model poisoning and adversarial threats.The resilience against attacks improved robust accuracy from 73.5% to 94.2%, while privacy budget adjustments achieved 93.3% accuracy at ε = 1.0.Compared to state of art leading models, EPP-CHBCFL demonstrates enhanced performance in security, scalability, and support for diverse edge devices, confirming its suitability for secure EHR management and real-time healthcare analytics.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Blockchain Technology Applications and Security Cryptography and Data Security

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
377–403
Publication date
23 March 2026
Language
EN