Artificial Intelligence and Machine Learning for Payroll Fraud Detection in the United States Public Sector Payroll Systems: A Scoping Review

Abstract

Payroll fraud remains a significant concern in United States public sector payroll systems because it may lead to financial loss, weaken accountability, and reduce public trust in public financial management. This scoping review maps available evidence on the application of artificial intelligence and machine learning to payroll fraud detection in United States public sector and related financial oversight contexts. The review was guided by the Joanna Briggs Institute framework and reported in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews. English-language studies published between 2016 and 2026 were searched across Scopus, Google Scholar, EBSCOhost, SpringerLink, and the Social Science Research Network. After screening and full-text assessment, 21 studies were included in the review. Thematic synthesis identified five main themes: the shift from manual and rule-based fraud detection to proactive AI- and machine learning-enabled monitoring; the importance of data quality, data integration, and system interoperability; the use of anomaly detection, risk scoring, alerts, and case prioritisation; the need for explainability, human oversight, privacy protection, fairness, and responsible AI governance; and institutional readiness, technical capacity, and the limited payroll-specific evidence base. The findings indicate that AI and machine learning can support earlier identification of payroll irregularities, including ghost employees, overtime abuse, duplicate payments, unauthorised salary changes, and suspicious disbursement patterns. However, the available evidence remains stronger in related areas such as banking, tax fraud, audit analytics, financial fraud detection, and payment integrity than in public sector payroll systems specifically. Effective implementation requires reliable data, interoperable systems, explainable models, skilled personnel, privacy safeguards, and human-in-the-loop decision-making.