An Enhanced Blockchain Federated Learning Approach for Privacy-Preserving and Secure Healthcare in Smart Hospitals

Ensuring security is a top concern nowadays in networks, and many approaches have been used, each with certain limitations. The existing methods, when used in a standalone setting, experience drawbacks: Classical Federated Learning (FL) relies on a centralized aggregator, which has limitations that raise the risk of manipulation and lack transparency, whereas other approaches that secure aggregation and Differential Privacy (DP) fail to verify client updates and reduce accuracy. Basic Blockchain-based FL uses consensus mechanisms and suffers from scalability, storage issues, and latency. Other approaches, reputation- and incentive-based, depend on static or manipulable measures, whereas aggregation methods, Krum or Median, perform poorly under non-IID data distributions. All these approaches focus on any of the measures such as privacy, robustness, or decentralization in isolation without providing a unified solution. Hence, the proposed enhanced BFL reference model provides a decentralized workflow with adaptive trust aggregation combines blockchain verified trust scoring, secure aggregation, attention-based weighting, and DP to improve robustness in heterogeneous settings. From experimental results of health hospitals, the proposed model achieves 98.40% accuracy and error rates of 0.80, over other methods of FedAvg (88%), FedProx (91%), and basic BFL (93%). Overall, the derived model integrates features such as adaptive trust, smart contracts, and secure aggregation, providing a more reliable and scalable solution for real-time decentralized environments along with reduced communication overhead, improved fault tolerance, and increased transparency as compared to other existing methods.