AN EXPLAINABLE AI FRAMEWORK FOR ETHICAL FRAUD PREVENTION IN U. S FEDERAL WELFARE PROGRAMS

Abstract

Fraud in U.S. federal welfare programs has continued to be a thorn in the flesh, and it is exacerbated by the disjointed data and the emergence of black box automated decision-making tools. The paper is based on a hybrid Explainable Artificial Intelligence (XAI) framework of machine-learning-based fraud detection, interpretable models, and ethical considerations, which are in alignment with federal AI governance principles. The system presents higher fraud detection, lower false-positive, and quantifiable fairness benefits using simulated eligibility, income, and transactional data based on SNAP workflows models, Medicaid workflows models, and TANF workflows models. Cross-agency data integration, including privacy preserving matching between SSA and IRS records, are also included in the study. The findings indicate that AI systems that are transparent and ethically regulated have the potential of making a significant enhancement in fraud prevention and safeguarding civil rights and enhancing public trust.