De-Biased Federated Learning Frameworks for Multi-Hospital Supply Chain Diagnostics and Localized Shortage Forecasting

• Billy Elly

  Lautech

  Author

Healthcare supply chains face unprecedented challenges from fragmented data systems, privacy constraints, and algorithmic biases that disproportionately affect underserved populations. Traditional centralized forecasting models fail to capture localized demand patterns while raising significant privacy concerns. This research develops and validates a de-biased federated learning framework for multi-hospital supply chain diagnostics and localized shortage forecasting. The proposed framework integrates fair aggregation mechanisms with performance-weighted model blending to address both statistical heterogeneity and systemic biases across participating institutions. Using a hybrid methodology combining retrospective electronic health record data from diverse hospital cohorts and prospective simulation, the framework achieved a 92.1% demand prediction accuracy  while reducing demographic-based prediction disparities by 38% compared to standard FedAvg baselines. The de-biased aggregation approach, incorporating fairness metrics into client weighting, maintained competitive predictive performance (89.4% of centralized model accuracy ) while ensuring local data sovereignty and regulatory compliance. Feature importance analysis identified length of stay, critical care admissions, and specialized procedures as primary drivers of supply demand . The framework provides a replicable, privacy-preserving solution for healthcare systems seeking equitable resource allocation and proactive shortage mitigation. Practical implications include reduced inventory holding costs (22%) and improved service levels during demand shocks (maintained above 80%) .

• PDF

Copyright (c) 2026 Billy Elly (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.