Artificial Intelligence-Assisted Financial Fraud Risk Assessment Using Explainable Ensemble Learning

Abstract

Financial fraud detection increasingly depends on machine learning models capable of identifying subtle irregularities across heterogeneous accounting, transactional, and behavioral data. This article proposes an AI-assisted fraud risk assessment framework that combines ensemble learning, feature-level explainability, and risk-oriented decision support. The framework integrates structured financial indicators, firm-level behavioral signals, and anomaly scores into a unified model designed to improve early detection of suspicious activities. Random forests, gradient boosting, and support vector machines are compared as base learners, while SHAP values are used to interpret model decisions and identify high-risk financial patterns. The study emphasizes the importance of balancing predictive accuracy with interpretability, particularly in auditing, banking supervision, and enterprise risk management contexts. By linking model outputs to explainable financial risk factors, the proposed approach supports transparent fraud screening and reduces dependence on black-box prediction. The article contributes to the literature by presenting a practical framework for explainable fraud risk assessment that can be adapted to financial reporting, credit evaluation, and supply-chain finance scenarios.