CLOSING THE TAX GAP: AN ANALYTICS-BASED FRAMEWORK FOR FRAUD-RISK DETECTION AND AUDIT PRIORITIZATION IN STATE AND FEDERAL TAX ADMINISTRATION

Abstract

The persistence of the tax gap illustrates the limitations of traditional compliance monitoring in the face of large-scale electronic filing, complex business structures, cross-jurisdictional activity and rapidly evolving fraud patterns. This study develops an analytics-based framework for fraud-risk detection and audit prioritisation for state and federal tax administration. A structured evidence synthesis and design-science approach was utilised on recent tax administration reports, peer reviewed tax analytics studies and explainable artificial intelligence literature with a focus on 2024-2026 evidence. The resulting framework leverages authorised federal and state data sources, data-quality controls, rule-based screening, anomaly detection, supervised machine learning, cross-jurisdictional consistency checks, explainable risk scoring, and model governance based on human review and feedback. We propose two decision measures: A Tax Risk Score that combines predictive probability, anomaly intensity, rule violations, compliance history, and cross-source inconsistency; and an Audit Priority Score that combines risk with expected recoverable revenue and administrative cost. The synthesis shows that predictive accuracy is not enough for public tax administration: model performance should be assessed in conjunction with precision–recall behaviour, audit yield, false-positive burden, explainability, fairness, privacy and model drift. Consequently, the framework positions analytics as a decision-support mechanism rather than an autonomous means of determining fraud. It offers a practical architecture for improving case selection while protecting taxpayer rights and directing scarce audit resources toward material noncompliance. Future validation should use de-identified administrative audit data that has been legally authorised across federal and state jurisdictions.