Machine Learning for Software Deployment in the Public Sector: A Systematic Review and Research Agenda for African Contexts

Failures in deploying digital public services can disrupt essential systems and affect millions of citizens. Machine learning (ML)-based software deployment decision support, including build risk prediction, release gating, autoscaling, rollback assistance, and post-deployment anomaly detection, offers opportunities for safer and more reliable releases. However, the extent of existing evidence in public-sector environments, particularly within African institutions, remains unclear. Following the PRISMA 2020 guidelines, this systematic literature review searched five databases using predefined inclusion criteria and a six-item quality assessment. A total of 33 peer-reviewed studies published between 2018 and 2025 were included, while studies focusing exclusively on MLOps were excluded. The findings reveal that none of the reviewed studies (0/33) explicitly evaluated ML deployment decision support in public-sector contexts or African institutions; existing evidence originates primarily from private-sector or unspecified environments. Research efforts are concentrated on autoscaling (12/33, 36%) and build prediction (9/33, 27%), with tree-based models being the dominant approach (16/33, 48%). Furthermore, only one study (3%) reported statistical significance testing or confidence intervals. This review identifies a research and evidence gap rather than confirming the absence of practical adoption. It proposes a staged research agenda toward explainable, lightweight, and context-aware ML deployment support for African public institutions.