基于非平衡大数据的公司破产评估模型研究  

Research on the Corporate Bankruptcy Evaluation Model Based on Unbalanced Data

作　　者：李田雨 高煌婷 翟亚琪 LI Tianyu;GAO Huangting;ZHAI Yaqi(Business School,Hubei University,Wuhan,Hubei 430062,China;Scool of Management,Wuhan University of Technology,Wuhan,Hubei 430070,China)

机构地区：[1]湖北大学商学院,湖北武汉430062 [2]武汉理工大学管理学院,湖北武汉430070

出　　处：《财经理论与实践》2025年第2期43-50,共8页The Theory and Practice of Finance and Economics

基　　金：湖北省高等学校哲学社会科学研究重大项目(23ZD179)。

摘　　要：大数据环境下,应用机器学习数据挖掘分析技术对波兰破产及未破产公司的财务数据进行建模训练和测试验证,其中包括多层感知器中的SMOTE、SMOTE-Borderline1和BMS不平衡算法。横向对比发现SMOTE、SMOTE-Borderline1、BMS算法有效提升了F1-Score,证明了多层感知器算法在公司破产评估领域内处理非平衡类别数据手段的有效性。纵向对比表明在不同的预测时间跨度上,MLP模型和公司财务数据的分类器模型效果具有显著差异。最后,使用卡方检验筛选出公司短期负债、资金结构和经营利润等较为重要的财务指标。Within the context of big data,this study employs machine learning and data mining techniques to model,train,and validate the financial data of bankrupt and non-bankrupt companies in Poland,including the application of SMOTE,SMOTE-Borderline1,and BMS,which are algorithms designed to handle imbalanced data within the multi-layer perceptron framework.A horizontal comparison revealed that the SMOTE,SMOTE-Borderline1,and BMS algorithms effectively enhanced the F1-Score,thereby substantiating the efficacy of multilayer perceptron algorithms in addressing imbalanced categorical data within the domain of corporate bankruptcy assessment.A vertical comparison indicated significant differences in the performance of the MLP model and the classifier model based on corporate financial data across various predictive periods.Ultimately,chi-squared tests were utilized to identify key financial indicators,such as short-term liabilities,capital structure,and operating profits,as being of considerable importance.

关 键 词：机器学习 多层感知器算法 破产评估模型 MLP模型 

分 类 号：F271[经济管理—企业管理] TP181[经济管理—国民经济]