An explainable machine learning model for consumer credit scoring in Mexico

Abstract

This paper proposes an explainable machine learning model for consumer credit scoring in Mexico, an emerging economy. We develop an extreme gradient boosting (XGBoost) model using non-traditional data from the Financial Inclusion National Survey. To address the black box problem, we explore the feature importance by estimating the Shapley values that measure the average marginal contributions across all possible subsets of features. The key drivers of consumer credit defaults include the adverse economic effects due to the COVID-19 pandemic and financial attitudes and behaviors. By exploring the distributions of the Shapley values by age and income, we find the evidence of non-linearity of the feature explanations.