An explainable credit scoring framework: A use case of addressing challenges in applied machine learning

Abstract

While Machine Learning (ML) classification algorithms can accurately classify a borrower's credit risk, the determinants of the credit score cannot be interpreted clearly by customers, decision makers and auditors. The lack of transparency of black-box credit scoring mechanisms reduces the trust in the banking system and has serious implications for the financing and growth of businesses. Recent regulations in the European Union and the United States require that credit decision mechanism should by explainable and transparent. We present a framework for developing an explainable credit scoring model. Our scientific novelty is to follow a simple and parsimonious Surrogate approach for credit scoring. This approach estimates an explainable white-box model that effectively fits to the in-sample forecasts of the most accurate 'black-box' model. We implement the Surrogate credit risk framework using check transactions data provided by a Turkish bank. We find that the Surrogate tree's performance is sufficiently close to performance of the most accurate black-box XGBoost model. Overall, our findings show that it is possible to develop a high-performing explainable credit scoring model with a minimal decrease in model accuracy.