Bankruptcy prediction with explainable machine learning methods

Ing. Munkhnaran Tsogoo

Master's thesis

Abstract:

The bankruptcy prediction model has been studied rapidly over the last few decades, and the methods used are becoming more sophisticated. As artificial intelligence and machine learning methods have evolved, using these methods in the bankruptcy prediction model has led to a higher prediction accuracy of default. Although highly predictable, machine learning methods to the use of the model in business are still lacking in explaining the reasons for the model. Therefore, this thesis aims to develop an explanatory machine learning method based on the 49,000 UK companies dataset. As a result of the study, the bankruptcy prediction model was developed using traditional and machine learning methods, and the model interpretations are finally compared. In other words, the scorecard is based on a logistic regression coefficient that was compared with the results of the XGBoost model results explained by the SHAP technique. We used the open-source R project (3.6) and Python (3.8) for all estimations.

Abstract:

The bankruptcy prediction model has been studied rapidly over the last few decades, and the methods used are becoming more sophisticated. As artificial intelligence and machine learning methods have evolved, using these methods in the bankruptcy prediction model has led to a higher prediction accuracy of default. Although highly predictable, machine learning methods to the use of the model in business are still lacking in explaining the reasons for the model. Therefore, this thesis aims to develop an explanatory machine learning method based on the 49,000 UK companies dataset. As a result of the study, the bankruptcy prediction model was developed using traditional and machine learning methods, and the model interpretations are finally compared. In other words, the scorecard is based on a logistic regression coefficient that was compared with the results of the XGBoost model results explained by the SHAP technique. We used the open-source R project (3.6) and Python (3.8) for all estimations.

Keywords

Bankruptcy prediction, credit scoring, machine learning, credit risk, logistic regression, XGBoost, scorecard, SHAP

Language used: English

Date on which the thesis was submitted / produced: 14. 5. 2021

Thesis defence

• Date of defence: 24. 6. 2021
• Supervisor: Ph.D. Oleg Deev
• Reader: Mgr. Zuzana Rakovská, Ph.D.

Masaryk University

Faculty of Economics and Administration

Master programme / field:
Finance / Finance