基于多阶段集成学习模型的企业破产风险评估研究

为了验证所提出的基于多阶段集成学习模型的企业破产风险评估模型的性能，本文一共使用了10个实验数据集进行验证。其中，9个数据集来自UC Irvine (UCI)机器学习存储库，它们分别是Australian，Japanese，German (Asuncion和Newman，2007)，Taiwan (Yeh和Lien，2009)和波兰破产企业数据集(Zięba等，2016)。波兰制造业企业破产风险数据集包含五个具有不同预测周期的独立数据集(即Polish 1，Polish 2，Polish 3，Polish 4和Polish 5)；1个数据集(Creator数据集)由中国电子政务服务提供商科创信息技术有限公司 于2019年发布，包含了35960家中国公司的产权，财务报表和基本公司信息。To verify the performance of the proposed multi-stage ensemble learning-based enterprise bankruptcy risk assessment model, a total of 10 experimental datasets were utilized for validation in this paper. Among them, 9 datasets are sourced from the UC Irvine (UCI) Machine Learning Repository, namely Australian, Japanese, German (Asuncion and Newman, 2007), Taiwan (Yeh and Lien, 2009), and the Polish bankruptcy enterprise dataset (Zięba et al., 2016). The Polish manufacturing enterprise bankruptcy risk dataset includes five independent datasets with distinct prediction horizons, i.e., Polish 1, Polish 2, Polish 3, Polish 4 and Polish 5. One additional dataset (Creator Dataset) was released in 2019 by Chuangxin Information Technology Co., Ltd., a Chinese e-government service provider, which contains the property rights, financial statements and basic corporate information of 35,960 Chinese companies.