A New Approach to Assessing the Sensitivity of a Credit Scoring Model

The banking system is a key sector of the economy and often faces various types of risks. This article focuses on credit risks, which are the main source of financial losses for credit institutions, including banks. Based on real data, a scoring model was developed in this study to assess and optimally manage credit risk, using the XGBoost machine learning algorithm. The model returns score values in the range of 300–850. Quality metrics such as ROC AUC, PR AUC, Precision, Recall, and F1 score were evaluated for the developed model. Within the scope of the research, a new approach was proposed to assess the sensitivity of a machine learning model to changes in variable values. The approach is based on the concept of entropy. The results show that the use of machine learning models makes it possible to assess and optimally manage credit risk, thereby reducing potential losses. Moreover, the proposed new approach, when applied together with the well-known SHAP method, allows for a deeper evaluation of the model and its sensitivity to changes in variable values.