H-VSAF: a verifiable and privacy-preserving federated learning framework for secure remote healthcare monitoring

The sensitive patient data handled by remote healthcare monitoring systems requires reliable, secure learning methods. Federated learning enables collaborative healthcare analytics without storing raw medical data centrally. However, existing techniques still suffer from issues including privacy leakage, untrusted aggregation, and frequent device outages. In this work, we propose H-VSAF, a privacy-preserving, verifiable, and lightweight federated learning system for healthcare environments. H-VSAF uses linear homomorphic hashing, Bloom filter-based online authentication, Shamir secret sharing, Diffie-Hellman key generation, and single-mask Secure Aggregation to ensure secrecy, precision, and resilience. The security analysis provides informal guarantees against gradient inference, aggregation tampering and deletion, and forged dropout claims under the threat model and cryptographic assumptions above. We experimentally evaluate H-VSAF on healthcare datasets and demonstrate that it achieves model accuracy within 0.2-0.8% of standard federated learning while decreasing client compute cost by up to 32% and communication overhead by 28% compared to traditional secure aggregation methods. Moreover, the framework is resilient to 40% client dropouts and introduces verification latency < 35 ms each round. The results clearly show that H-VSAF provides a credible and practical platform for safe joint health care analytics.