Examining the Research Gap in Federated Learning for Privacy-Preserving Healthcare Data Analytics in Nigerian Hospitals

Machine learning applied to healthcare data in Nigerian hospitals could significantly improve diagnostic accuracy, resource allocation, and epidemiological surveillance, yet patient data privacy concerns and data siloing across hospital systems prevent centralised dataset creation, a research gap that federated learning frameworks could address. This study examined the research gap in federated learning for privacy-preserving healthcare analytics in Nigerian hospital contexts through systematic review and a proof-of-concept empirical evaluation. A systematic literature review of federated learning healthcare applications published from 2020 to 2024 identified 67 publications, of which only 3 included data or context from sub-Saharan African hospitals and none from Nigerian institutions. A proof-of-concept federated learning system using PySyft was implemented simulating three Nigerian teaching hospitals (Lagos, Ibadan, and Abuja) as federated nodes, each holding a de-identified subset of a chest X-ray pneumonia classification dataset partitioned to simulate hospital-specific data distributions. A federated ResNet-18 model trained across three nodes for 50 rounds achieved 87.3 percent classification accuracy on a shared validation set, compared to 89.1 percent for a centralised baseline trained on the combined dataset. Communication overhead per round was 42 MB per participating node. The accuracy gap was 1.8 percentage points, consistent with theoretical expectations for non-IID federated data. The study identifies infrastructure connectivity, computational requirements for resource-constrained hospitals, and the absence of Nigerian healthcare federated learning datasets as the three primary research gaps requiring priority attention.

Keywords: federated learning, healthcare analytics, privacy-preserving, Nigerian hospitals, distributed machine learning