Explainable Artificial Intelligence (XAI) in the Banking Sector: A Review of Techniques for Transparent Credit, Fraud and Compliance Decisions (Advances in Economics Management and Political Sciences)

Summary

• Banks use complex machine learning models such as neural networks, random forests and gradient boosting for credit scoring, loan approval, fraud detection, risk assessment and anti-money laundering, and many operate as black boxes.p.1
• The review covers inherently interpretable models, post-hoc techniques such as LIME and SHAP, and counterfactual explanations that tell customers what would change a decision.p.1
• Explanations can be global or local, and can be feature attributions, example-based or counterfactual.p.2
• Equality of opportunity requires equal true-positive rates across groups, offering a measurable fairness criterion for lending.p.2
• The paper proposes a layered XAI architecture in which explanation and governance are integral parts of the banking decision pipeline.p.3
• The EU AI Act classifies credit scoring as high-risk.p.4

Key insights

1. Explanations from different methods can disagree, and there is no standard for choosing among them.p.3
2. The stability and faithfulness of explanations are rarely validated in production banking systems.p.3
3. Explanations designed for data scientists are often unsuitable for customers or branch staff.p.3
4. Fraud and anti-money-laundering models raise explanation challenges distinct from credit scoring and are less studied.p.3
5. In the proposed design an interpretable scorecard serves as a benchmark and fallback alongside a high-performance model such as gradient boosting.p.3
6. Customers should receive plain-language reasons and steps for improvement, while auditors and regulators access full documentation and logs.p.4
7. SHAP and TreeSHAP are rated high for stability and suit credit and fraud models, while LIME is rated low to medium.p.4
8. Future work includes standardized explanation quality metrics and privacy-preserving explanations.p.4