Explainable Artificial Intelligence for Financial Risk Management: Integrating Fraud Detection, Credit Scoring, and Regulatory Compliance

Explainable Artificial Intelligence for Financial Risk Management: Integrating Fraud Detection, Credit Scoring, and Regulatory Compliance

Mehrad Farrahi1

1) Faculty of Computer and Information Technology Engineering, Islamic Azad University, Tehran, Iran

IOI Code : ABGZ-ACADKA
Publication : The third international conference on new approaches in engineering and basic sciences(icnabs.ir)
Abstract :
    Background: Artificial intelligence is widely used to detect suspicious transactions, estimate creditworthiness, and support financial compliance; however, opacity, data instability, potential discrimination, and third-party dependence complicate deployment in regulated settings. Explainable artificial intelligence (XAI) promises understandable model outputs, but explanations do not automatically establish causal validity, fairness, or legal compliance. Objective: This article integrates recent evidence on explanation methods, financial decision tasks, and regulatory expectations into an operational framework for risk-sensitive financial institutions. Methods: A structured, non-exhaustive integrative review of research and authoritative policy publications from 2024–2026 was conducted using targeted scholarly and regulatory searches. Evidence was organized by use case, explanation objective, technical validity, stakeholder, and governance control; no patient, consumer, transaction-level, or experimental data were collected. Results: The synthesis identifies five recurring tensions: prediction versus explanation fidelity; rare-event fraud detection versus representative explanation baselines; individual credit reasons versus group fairness; model transparency versus confidentiality and adversarial risk; and regulatory documentation versus continuous, real-world monitoring. A seven-layer architecture and a stage-gated evaluation protocol are proposed, with distinct controls for transaction fraud, consumer credit, and compliance monitoring. Discussion: XAI should be managed as a lifecycle assurance capability rather than a visualization attached to an opaque classifier. Risk-aware model selection, faithful and stable explanations, calibrated performance, fair treatment, human challenge, and reconstructable evidence records are complementary requirements. Conclusion: Integrated financial XAI can improve decision accountability when deployment is tied to independently validated explanations and context-specific legal duties. Prospective, multi-institutional evaluation is needed before quantitative gains can be claimed.

Keywords : explainable artificial intelligence financial risk management fraud detection credit scoring model risk governance shap counterfactual explanations algorithmic fairness eu ai act regulatory compliance