Model selection for predicting the default probability of individual unsecured loans: The case of Asia Commercial Joint Stock Bank (ACB)

Abstract

This study was conducted to select a predictive model for the probability of default within an internal credit rating system for individual unsecured loans. Consequently, the study also suggests integrating the model into the credit process and proposes the collection of accurate information to enhance the model’s predictive quality. In this research, the authors used a dataset sampled from January 01, 2022, to December 31, 2022, and observed from January 01, 2023, to December 31, 2023. We utilized nine Machine-Learning models and an Ensemble model. We also implemented data balancing and preprocessing features before model estimation. The Logistic and Ensemble models were the two best predictive models. With an optimal bankruptcy probability threshold for sensitivity prediction, the Logistic model performed better than the Ensemble. Additionally, the analysis revealed the importance of predictive variables, including Educational level, Organization type, Gender, Age, Continuous income duration, Employment duration, Loan term, Borrowing needs, Total income, Total expenses, Monthly debt obligations, and Credit history for the last six months.