Forecasting the Customer's Solvency for Issuing a Car Loan Based on an Analysis of the Characteristics of the Desired Car

This study is aimed at improving the prediction of the customer's solvency when issuing a car loan by analyzing the characteristics of the desired car. Traditional credit scoring methods often do not allow you to cover the full range of risk factors associated with non-repayment of a loan. This research aims to improve the accuracy of predicting solvency by integrating vehicle-specific data such as make, model, year of manufacture, mileage, price and engine type into machine learning models. This article explores the application of various machine learning methods to a dataset provided by a large Russian bank. The goal is to determine the best model that gives the most accurate forecast for the target variable - the customer's solvency within 12 months. Nine machine learning methods were applied, and their effectiveness was compared using ROC-AUC and GINI indicators. The highest GINI score - 0.39 - was obtained using the ensemble, stacking model.