Conceptual Framework for Automated and Self-Improving Credit Risk Modeling
Conceptual Framework for Automated and Self-Improving Credit Risk Modeling

Summary/Abstract: Credit risk scorecard development remains a largely expert-driven process, despite the growing complexity of financial data and the increasing availability of automated modeling tools. Existing approaches often focus on model selection or pipeline optimization, but provide limited support for systematic knowledge reuse and self-improvement across repeated modeling tasks. This paper proposes a conceptual framework for automated and self-improving credit risk modeling pipeline generation. The framework is designed to support the construction of end-to-end modeling pipelines through reusable experiment history, warm-starting, and decision-based pipeline construction. Its main contribution is the positioning of credit risk model development as a learning process in which previous modeling experience informs future pipeline decisions.As a conceptual study, the paper does not present empirical validation, but establishes a methodological foundation for future prototype implementation and evaluation against expert-driven scorecard development and existing AutoML systems.

• Details
• Contents