A novel multi-stage ensemble model with fuzzy-clustering and optimized classifier composition for corporate bankruptcy prediction

In this experiment, the datasets are from the UC Irvine (UCI) UCI machine learning repository (Zięba et al., 2016), which contains the financial indicators of Polish manufacturing corporates from 2007 to 2011 in the real world. The datasets were separated into five parts (each part represents each fiscal year) that describe the period from the 1st year (2007 fiscal year) to the 5th year (2011 fiscal year), which corresponds to five different bankruptcy cycles. The class labels (“0” is operating and “1” is bankruptcy) of the datasets are determined by the bankruptcy status of the enterprise in 2012. Furthermore, the Creator dataset from the real world that was published by a Chinese intelligent government services provider called Creator Information Technology Co., Ltd in 2019 was also adopted. The Creator dataset includes company management information of 35960 Chinese companies.

本实验所用数据集源自加州大学欧文分校（UC Irvine）机器学习库（Zięba等人，2016），其中包含2007至2011年波兰制造业企业的真实经营财务指标。该数据集被划分为五个子集，每个子集对应一个财年，覆盖2007财年（第1年）至2011财年（第5年）的时段，对应五个不同的破产周期。数据集的类别标签（“0”代表正常经营，“1”代表破产）以企业2012年的破产状态作为判定依据。此外，本实验还采用了2019年由中国智能政务服务提供商创兆信息技术有限公司（Creator Information Technology Co., Ltd）发布的真实场景Creator数据集，该数据集涵盖35960家中国企业的经营管理信息。