Entropy method of constructing a combined model for improving loan default prediction: A case study in China

Abstract

In recent years, credit scoring has become an efficient tool to assist financial institutions in identifying potential default borrowers, and the combined model is widely viewed as a useful vehicle. In this study, after pre-processing based on random forest, we propose a combined logistic regression algorithm and artificial neural network model to improve the predictive performance based on actual data from a rural commercial bank under the condition that loan quality directly affects the profitability of the bank. The combined model requires a step with an entropy method to determine the entropy weights of the logistic regression model and artificial neural network model. The experimental results reveal that the proposed combined model outperforms the two base models on four evaluation metrics: accuracy (ACC), area under the curve (AUC), Kolmogorov-Smirnov statistic (KS), and Brier score (BS). Moreover, the model is superior to a state-of-the-art ensemble model, stacking.

Keywords:

• Credit scoring
• combined model
• entropy method
• stacking

Acknowledgements

The author would like to thank the editors and anonymous reviewers for their helpful comments and suggestions.

Disclosure statement

The author declares no conflict of interest.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China [No. 71373173], National Social Science Found of China [No. 14ZDB135], the Major Research Plan of Social Science Found of Tianjin Municipal Education Commission [No. 2019JWZD39] and the Major Research Plan of Social Science Found of Tianjin [No. TJGL7-016].