Federated Machine Learning for Privacy-Preserving Health Data Analysis Across Distributed Nigerian Hospital Networks

Health data governance in Nigeria is characterised by institutional fragmentation, limited interoperability between hospital information systems, and significant patient privacy concerns that collectively obstruct the large-scale data analysis needed to generate actionable clinical and epidemiological insights. Federated machine learning, which enables model training across distributed data sources without centralising sensitive patient records, offers a promising architectural solution to these governance constraints. This study develops and evaluates federated machine learning protocols specifically designed for the distributed hospital network environment of Nigeria, addressing the unique challenges posed by heterogeneous data formats, variable network reliability, and unequal computational capacities across public and private hospitals. The research is grounded in a collaborating network of eight hospital information technology teams across Lagos, Abuja, Enugu, and Kano, providing access to de-identified model training datasets in compliance with the Nigeria Data Protection Act 2023 and the National Health Act 2014. Novel contributions include a Bandwidth-Adaptive Federated Averaging Protocol designed to maintain model quality during connectivity fluctuations common in Nigerian hospital network environments, and a Differential Privacy calibration methodology adapted for the smaller patient population sizes characteristic of Nigerian specialist hospitals. Federated models for diabetes complication risk prediction and malaria treatment outcome classification are evaluated, with the federated approach achieving model accuracy within 2.8 percent of centralised training baselines while providing formal privacy guarantees. Keywords: federated machine learning, health data privacy, Nigerian hospitals, differential privacy, distributed learning.