ProsperUkachi/Final-year-project-public

Reproducible implementation accompanying the study "Development of a Federated Learning Model for Fair Healthcare Data Analysis in Selected African Countries." # Federated Learning for Fair Healthcare Data Analysis ## Overview This repository contains the implementation and supporting materials for the final-year research project: > **Development of a Federated Learning Model for Fair Healthcare Data Analysis in Selected African Countries** The project investigates whether federated learning can support healthcare prediction across geographically separated datasets while preserving data locality, and examines how federated learning failure conditions can affect predictive performance, algorithmic fairness, and the stability of model explanations. The study uses Demographic and Health Surveys (DHS) microdata from **Nigeria DHS 2024** and **Kenya DHS 2022** to predict adequate antenatal care (ANC) utilisation. A simulated horizontal federated learning environment is created in which data remain separated across four clients representing country and urban-rural strata. The experiments compare conventional Federated Averaging (FedAvg) with coordinate-wise median aggregation under normal and adverse federated learning conditions. Centralised models are also used as reference baselines. ## Research Questions The implementation addresses the following broad questions: 1. How does federated learning perform for ANC utilisation prediction when healthcare data are distributed across heterogeneous clients? 2. How do client failures, stale updates, and Byzantine-corrupted updates affect predictive performance and fairness? 3. Can robust aggregation reduce the effect of malicious or anomalous client updates? 4. How stable are SHAP-based model explanations under federated learning compared with a centralised reference model? ## Data The project uses DHS microdata from **Nigeria DHS 2024** and **Kenya DHS 2022**. The DHS datasets are **not included in this repository** because they are distributed under the terms and access conditions of The DHS Program. Researchers attempting to reproduce the study should obtain the relevant …