Data analytic approach for bankruptcy prediction. (30th December 2019)

• Record Type: Journal Article
• Authors: Son, H.; Hyun, C.; Phan, D.; Hwang, H.J.
• Is Part Of: Expert systems with applications. Volume 138(2019)
• Publication Date: 2019-12-30
• Subjects: Bankruptcy prediction -- Data analysis -- Machine learning -- Boosting -- Preprocessing -- Box–Cox transformation -- Feature importance; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2019.07.033
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: Features to predict a company's bankruptcy have highly skewed distributions. A Box–Cox transformation is a powerful technique to remove skewness of data. Machine learning algorithms are more suitable for bankruptcy prediction than statistical models. By combining feature's importance, we can understand the model. Graphical abstract: Abstract: Bankruptcy prediction problem has been intensively studied over the past decades. From traditional statistical models to state of the art machine learning models, various predictive models are developed and applied to various datasets. However, models that use machine learning are not used in the field of business, for two main reasons. First, the prediction accuracy does not far exceed the statistical models and second, the results are not interpretable. In this study, we focused on solving the skewness which is a characteristic of financial data. By solving this problem, we obtained 17% average improvement in AUC over existing models. To address the second shortcoming, we analyze the importance of features identified by the XGBoost model. The interpretation of the model differs among categories of data. Our bankruptcy prediction model has high predictive accuracy with clear explanations and is therefore directly applicable to the industry.