AI-Driven Anomaly Detection for Public Expenditure: A Prospective Framework for Strengthening Oversight of Irregularities, Waste, and Potential Fraud through Explainable and Generative AI

Abstract

Public expenditure systems are essential to state capacity and public trust but remain vulnerable to errors, weak competition, inflated contracts, collusive behavior, and other indicators of waste or potential fraud. Traditional oversight is often retrospective, manual, and limited by fragmented data and rigid rule-based checks. This manuscript proposes a forward-looking AI framework for anomaly detection that integrates machine learning, graph analytics, natural language processing, and a retrieval-grounded Generative AI explanation layer to support audit triage.The framework consolidates structured transactions, procurement records, contracts, invoices, vendor data, and project information into a unified analytical architecture. Statistical anomaly models, graph-based relational indicators, and document intelligence surface risk signals, while a controlled Generative AI component converts evidence-linked outputs into auditor-readable summaries without replacing human judgment.Rather than reporting implementation results, the manuscript provides a research and deployment blueprint. It defines a taxonomy separating anomalies, errors, irregularities, waste, abuse, collusion, corruption, and fraud; outlines a multimodal scoring pipeline; proposes evaluation criteria; and specifies governance safeguards for responsible, legally defensible use. A human-centered, explainable, governance-aware AI framework may strengthen oversight, subject to validation in pilot settings.