Federated Blood Supply Chain Demand Forecasting: A Case Study

Hanzhe Wei, Na Li, Jiajun Wu, Jiayu Zhou, STEVE DREW

Keywords: Federated Learning, Blood Supply Chain, Demand Forecasting

TL;DR: We explore the effectiveness of leveraging FL to enhance blood supply chain demand forecasting.

Abstract: Blood transfusion is a commonly used, life-saving medical therapeutics worldwide. A significant challenge is the high variability of supply and demand in blood products, making it difficult to maintain a balance between preventing shortages of blood products and preventing wastage. Recent studies used data-driven methods on demand forecasting for blood products from regional and centralized databases due to regulatory restrictions, which lack the panorama view of the national blood supply and demand picture. Motivated by achieving better policy-making through national blood supply chain demand forecasting, in this paper, we propose to use federated learning (FL) to forecast the demand for platelets through a case study with simulated scenarios considering a national demand and supply network. Our solution facilitates FL with a Long-Short-Term Memory (LSTM) model to make collaborative predictions for future decision-making processes from distributed and regional time-series data. Empirical studies show that FL brings additional performance improvement in various settings, especially for regions with scarcer and shorter data histories. We release the source code for our study at https://github.com/denoslab/fl-blood-supply-chain.