Better credit decisioning through scorecard surrogate models for machine learning algorithms

Over the last several years, the application of machine-learning models, called black-box models, has become a popular research topic in credit scoring. This study illustrates how surrogate models can be used to interpret credit decisions made using black-box models. A framework for using surrogate models in a credit scoring context is used to explain and interpret well-known machine learning models (e.g., neural networks, forests, gradient boosting, and support vector machines). This study uses real-world anonymised consumer bureau data obtained from Equifax to illustrate the degree of interpretability that can be achieved using machine learning models to assess the creditworthiness of loan applicants. The main objective of this study is to show practitioners how surrogate scorecard models can be used to interpret some of the most popular machine learning models in a credit scoring decision making process.