Predicting default risk for small and medium enterprises in Vietnam using machine learning models

Abstract

This study develops a model for predicting default risk (DR) for small and medium-sized
enterprises (SMEs) in Vietnam using machine learning methods such as Logistic Regression (LR), Decision
Trees, XGBoost, and Artificial Neural Networks (ANN). The data is collected from the financial statements
of enterprises borrowing from commercial banks and companies listed on the Vietnamese financial market
from 2010 to 2022. The performance of the models is evaluated using metrics such as the F1 score and
accuracy (ACC). Results show that Decision Trees, XGBoost, and ANN outperform LR. Specifically, ANN
achieves an F1 score of 0.756 and an ACC of 0.9345 on the validation dataset, demonstrating excellent
predictive capability. The ANN method has significant potential in identifying high-risk customers, thereby
optimizing the credit risk management process. The study also identifies key predictive variables, providing
insights for developing more effective DR models. Future research could apply advanced hyperparameter
tuning techniques and expand the feature set to optimize the model further.