Year: 2025
Type: article
Abstract: This article explores the architectural approaches for building explainable artificial intelligence (XAI) systems specifically designed for payment compliance testing in regulated financial environments. As financial institutions increasingly adopt sophisticated machine learning models to enhance compliance verification, they face the challenge of balancing advanced detection capabilities with regulatory requirements for transparency and explainability. The article examines the "black box" problem inherent in neural networks and proposes decision-tree surrogate models as a practical solution to bridge the interpretability gap. It further explores the implementation of SHAP values to quantify feature importance in payment decisions, providing crucial transparency for compliance officers and regulators. The article addresses regulatory considerations for XAI deployment, highlighting the need for comprehensive ML governance frameworks that include robust documentation, stakeholder-appropriate explanations, and rigorous testing methodologies. Finally, it presents an implementation architecture that preserves explainability throughout the transaction lifecycle, demonstrating how financial institutions can satisfy both performance and transparency requirements in payment compliance systems.
Author: Aparna Thakur
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4409675862