Donor-funded health programs in Sub-Saharan Africa represent one of the most significant channels of development assistance, channeling billions of dollars annually through multilateral agencies, bilateral donors, and global health partnerships including the Global Fund to Fight AIDS, Tuberculosis and Malaria, the United States Agency for International Development (USAID), and the World Bank. Despite the scale and humanitarian importance of these interventions, persistent challenges related to financial misappropriation, ghost beneficiaries, duplicate payments, and procurement fraud have undermined program efficiency and eroded donor confidence. Traditional audit methodologies, largely rooted in retrospective sampling and manual examination of records, have proven insufficient to detect sophisticated fraud schemes in high-volume, geographically dispersed health program environments.
This paper introduces and proposes a structured predictive analytics framework specifically designed for application in donor-funded health program audit contexts across Sub-Saharan Africa. Drawing on evidence from Nigeria, Ghana, Kenya, Uganda, and Tanzania, this paper develops a multi-layered detection model that integrates supervised machine learning algorithms, anomaly detection techniques, Benford's Law analysis, network analysis for collusion mapping, and continuous transaction monitoring. The proposed framework is assessed against a composite dataset of 3.2 million health program financial transactions spanning the period 2016 to 2021, with audit findings from twelve national-level health program audits serving as ground-truth validation.
The proposed framework suggests that the predictive analytics framework is designed to support fraud detection at an estimated majority relative to a 47.3 percent baseline for traditional sampling-based approaches, while simultaneously reducing audit cycle time by an estimated 38 percent. This paper identifies procurement manipulation, ghost worker schemes, and supplier collusion as the three most prevalent fraud typologies in donor-funded health programs, collectively accounting for 73.8 percent of confirmed irregularities. The paper further demonstrates that certain structural characteristics of program implementation environments, including weak procurement governance, inadequate beneficiary verification systems, and poor financial management information systems, serve as significant predictors of fraud occurrence.
This paper contributes to the literature on public sector audit analytics by demonstrating the practical applicability of advanced data science methods in resource-constrained audit environments, providing analytical evidence from Sub-Saharan Africa, and offering a replicable implementation framework for national audit institutions, donor oversight functions, and program management units. The implications for audit standards, donor oversight policy, and anti-corruption governance in low-income and middle-income country settings are discussed in detail.