Shapley-Value based Contribution Measurement for Federated Learning on Blood Cells

Federated Learning (FL) enables decentralized machine learning across multiple entities without sharing centralized datasets, making it crucial for privacy-sensitive domains like healthcare. However, the lack of fair compensation mechanisms can deter data contributions. This paper introduces a Shapley Value-based Contribution Measurement (SVCM) method for FL on blood cells, using cooperative game theory to ensure fair and accurate reward distribution. Compared to traditional Leave- One-Out (LOO) methods, SVCM offers significant improvements in fairness and computational efficiency. Experimental results indicate that SVCM effectively measures each client's contribution in imbalanced data scenarios, ensuring fair compensation and promoting sustainable collaborative learning. This advancement can enhance medical research and patient care by encouraging broader adoption of federated learning in healthcare.