Modeling automobile credit scoring using machine learning models

Abstract

This study aimed to apply ML models to determine variables related to automobile credit scoring in a financial institute. We used four machine learning methods, including logistic regression (L.R), artificial neural network (ANN), random forest (R.F.), and Extreme Gradient Boosting (Xgboost) with 10-fold cross-validation repeated 10 times. The RF algorithm achieved the best performance in all performance metrics. It performed 96.2% under the receiver operating characteristic (AUROC) score. AUROC scores for other models were 91.7% for LR, 91.6% for Xgboost, and 91.5% for ANN. SHAP (SHapley Additive exPlanations) values were also calculated to better explain the indicators’ importance. The most relevant features of the model were debt to income ratio score, documented wealth score, and down payment rate score. To sum up, this study might help automobile credit providers and applicants for their credit evaluation process.