Weighted Federated Learning with Encryption for Diabetes Classification

📝 Abstract

This paper proposes a Weighted Federated Learning framework with Encryption (WFLE) for diabetes classification across distributed clinical sites. Patient privacy regulations and institutional data governance policies often prevent direct data sharing, motivating federated approaches. The proposed framework introduces a performance-based weighting scheme for model aggregation that down-weights contributions from sites with lower data quality or smaller sample sizes. Homomorphic encryption is integrated into the aggregation protocol to prevent inference attacks on model updates. Experiments on simulated federated scenarios demonstrate that WFLE achieves superior diabetes classification performance compared to standard FedAvg under heterogeneous data distributions, while maintaining strong privacy guarantees.

📋 BibTeX Citation

@inproceedings{zhao2025weightedfl,
  title     = {Weighted Federated Learning with Encryption 
               for Diabetes Classification},
  author    = {Zhao, Puyang and Yue, Z. and Liu, X. and Wu, J.},
  booktitle = {2025 Second International Conference on Artificial 
               Intelligence for Medicine, Health and Care (AIxMHC)},
  year      = {2025},
  publisher = {IEEE}
}

🔗 Related Publications