Explainable AI with Capsule Networks for Credit Risk Assessment in Financial Systems

This study introduces an Explainable AI (XAI) framework leveraging Capsule Networks (CapsNet) for credit risk assessment in financial systems. Traditional credit risk models often prioritize predictive accuracy over interpretability, creating challenges in stakeholder trust and regulatory compliance. CapsNet addresses this by maintaining spatial relationships in data while offering robust predictive capabilities. The proposed framework integrates data preprocessing techniques such as handling missing values and balancing imbalanced datasets, ensuring reliable inputs for CapsNet. Explainability is enhanced through Capsule Activation Vectors and attention mechanisms, elucidating the influence of critical features like income, loan amount, and credit history on predictions. The study employs SHAP values to identify key determinants of loan default, with income emerging as the most significant predictor. Experimental results validate CapsNet's superior performance, achieving 90% accuracy and F1-score, outperforming models like Logistic Regression and Random Forest. Additionally, the framework provides interpretative insights into model decisions, fostering transparency and trust. This work represents a significant step towards integrating explainable AI into credit risk management, ensuring compliance with financial regulations while improving operational efficiency and decision-making in the banking sector. By addressing the dual challenges of accuracy and interpretability, this research demonstrates how XAI frameworks can transform financial systems, supporting ethical and knowledge-based practices.