Privacy-Preserving Healthcare Data Analytics Using Blockchain and Federated Learning: The PrivaHealth FL System

Abstract

This paper addresses the escalating challenges of data privacy and security in healthcare analytics, particularly as medical institutions increasingly depend on big data and collaborative research.PrivaHealth FL, a comprehensive system that merges Federated Learning (FL) with Blockchain technology, is introduced to facilitate secure, privacy-preserving analysis of distributed medical data.Unlike prior frameworks that address individual security concerns in isolation, PrivaHealth FL integrates three orthogonal defense mechanisms: Differential Privacy (DP) using the Gaussian Mechanism with adjustable privacy budget ε ∈ [0.1, 2.0]; Homomorphic Encryption (HE) to protect gradient confidentiality during aggregation; and Byzantine Fault Tolerance (BFT) via a Krum-style distance filter to neutralize malicious participant updates.The system is built on a custom blockchain ledger providing full auditability through SHA-256 linked blocks and ECDSA digital signatures.Experimental evaluation across five simulated hospitals demonstrates that PrivaHealth FL achieves 86.8% global accuracy after 20 federated rounds -only 3.3% below an unprotected baseline -while reducing membership inference attack success rates by 87.3% at ε = 0.5.Byzantine fault detection achieves a 100% true positive rate with 0% false positives.A formal Threat Model analysis covering seven attack vectors confirms that PrivaHealth FL provides comprehensive, multi-layered protection suitable for realworld clinical deployment.