A Blockchain-Driven Federated Learning Framework for Robust Healthcare Systems: Overcoming Data Heterogeneity and Ensuring Auditability

Abstract

General Background: Automated Decision-Making Systems in healthcare require access to distributed medical records while maintaining strict patient data privacy compliance. Specific Background: Federated learning enables collaborative training across decentralized health institutions but faces vulnerabilities regarding data heterogeneity and lack of verifiable auditability. Knowledge Gap: Existing decentralized frameworks struggle to handle non-identically distributed datasets and malicious updates without introducing centralized aggregation dependencies. Aims: This study evaluates a reputation-weighted federated learning framework integrated with permissioned blockchain governance to enhance model accuracy and operational auditability. Results: Experimental evaluations on public medical benchmarks demonstrate an accuracy of 84.2 percent while maintaining over 80 percent accuracy under 10 percent data poisoning attacks. Novelty: The framework combines smart contract Service Level Agreement verification with dynamic on-chain reputation scoring to filter low-quality updates. Implications: Healthcare networks can deploy privacy-preserving artificial intelligence platforms that ensure regulatory compliance and robust model aggregation. Keywords: Federated Learning, Blockchain Governance, Healthcare AI, Data Heterogeneity, Smart Contracts Key Findings Highlights Dynamic reputation scoring effectively filters malicious updates and preserves global model stability during data poisoning attacks. Permissioned smart contracts automate governance and enforce service level agreements across distributed clinical nodes. Off-chain clinical record management ensures strict alignment with data protection regulations and right to erasure requirements.