Stakeholder centric approach to applying machine learning to probability of default models
Details
- Authors: Dyon Kok
- Title: Stakeholder-centric approach to applying machine learning to probability of default models
- Supervisor: Dr. Jörg Osterrieder, Dr. Marcos Machado
- Degree: Master of science
- University: University of Twente
- Year: 2024
- Status: Master thesis
Abstract
This thesis explores the integration of Explainable Machine Learning (XML) techniques, specifically the Explainable Boosting Machine (EBM) and Generalized Additive Models with Interactions Network (GAMINet), within the framework of Probability of Default (PD) modeling in financial institutions. Despite the advanced capabilities of EBM and GAMINet, our comparative analysis revealed that the performance of these newly implemented models did not surpass the existing system. However, the study underscores the potential of XML to enhance PD modeling by balancing predictive accuracy with interpretability, thus fostering stakeholder trust. We evaluated the models based on several criteria essential for interpretable machine learning applications, including explainability, regulatory compliance, and operational feasibility. Our findings suggest that while both models exhibit high levels of explainability and transparency, the choice between EBM and GAMINet should be guided by the specific needs and requirements of stakeholders. The thesis concludes with recommendations for integrating XML techniques into PD modeling practices and suggests avenues for future research, such as the seamless integration of Python packages into SAS environments and the exploration of advanced feature transformation techniques. This research contributes to the ongoing dialogue on leveraging machine learning advancements within the regulatory and ethical constraints of the finance sector, aiming to improve model performance and stakeholder engagement.
Keywords: Credit Risk Modeling, Machine Learning, Financial Data Analysis, Probability of Default, GamiNet, EBM, stakeholders