Machine Learning Application In Credit Scoring For Vietnam's Retail Loan

Abstract

The role of credit scoring in lending decisions can not be overemphasized for financial institutions and the economy at large. An accurate and well-performing credit scorecard allows lenders to control their risk exposure through selective credit allocation, based on historical customer data statistical analysis. We illustrate the improvement in model performance arising from the 2-stage model idea by applying it for a real-world credit dataset of one Vietnamese commercial bank. Due to commercial sensitivities surrounding the use of credit data, very few empirical studies which directly address this topic are published in Vietnam. This thesis using the Gini ratio as a core performance measure for the quantitative comparison of seven models including two 2-stage models and five traditional models in credit scoring.