Affiliations: [a] College of Computer Science and Technology (Software College), Henan Polytechnic University, Jiaozuo, Henan, China | [b] Safety and Emergency Management Research Center, Henan Polytechnic University, Jiaozuo, Henan, China | [c] Institute of Policy and Management, Chinese Academy of Sciences, Beijing, China | [d] CISUC, Department of Informatics Engineering, University of Coimbra, Coimbra, Portugal
Correspondence:
[*]
Corresponding author: Ning Chen, College of Computer Science and Technology (Software College), Henan Polytechnic University, 2001 Century Avenue, Jiaozuo 454003, Henan, China. E-mail:nchenyx@outlook.com
Abstract: It is well known that bankruptcy patterns are different across the industries, and consequently most studies focused on a given business sector or sub-sector. However, in some real world applications the bankruptcy patterns are likely constructed based on the companies of various business sectors (e.g. construction, real estate, education, retail, transportation) due to the lack of default samples. This paper presents a comparative study using different classifiers and performance metrics to examine the role of the business sector in corporate failure prediction based on the data from multiple business sectors. The experiments use a real-world French database of corporate companies diversified in different industries. The bankruptcy of companies is forecasted using 10 prediction models and evaluated by 8 performance metrics.
The experimental results are analyzed by means of multidimensional scaling, which transforms the high dimensional metric data into a 2D space through a pairwise distance preserving projection. Through the interpretation of the results, we can find that the business sector used as a predictor significantly enhances the exploratory power of prediction models. Moreover, by analyzing the results there is evidence that not only tangible benefits are attained with business sector information but also the advanced ensemble approaches yield good to excellent improvements in this financial setting.
Keywords: Bankruptcy prediction, business sector, multidimensional scaling, performance metric
