bankruptcy-early-warning-data
收藏资源简介:
该数据集为波兰公司破产预警系统的处理表格,源自UCI机器学习库的波兰公司破产数据(数据集365)。原始数据为3year.arff文件,经过解析为CSV格式,并删除了不含可用比率的行。未对数值进行插补或转换,因此本数据集为原始分析表。数据集包含10,503条公司-年度记录,共65列(64个财务比率加目标变量)。目标变量为bankrupt,表示该公司在3年内申请破产。正例样本数为495(占比4.71%),缺失单元格总数为9,888。数据字典文件(data_dictionary.csv)提供了每个比率的财务定义、类别和缺失率。注意事项:数据基于波兰申报,破产企业覆盖2000-2012年,存活企业覆盖2007-2013年,绝对违约率仅适用于该群体,不应视为其他市场的当前比率。该数据集适用于表格分类任务,如破产预测、信用风险评估等。
This dataset is a processed table for a Polish company bankruptcy early warning system, derived from the UCI Machine Learning Repositorys Polish companies bankruptcy data (Dataset 365). The original data is a 3year.arff file, parsed into CSV format, and rows without available ratios have been removed. No imputation or transformation of values has been performed, so this dataset is the raw analysis table. The dataset contains 10,503 company-year records with 65 columns (64 financial ratios plus the target variable). The target variable is bankrupt, indicating whether the company filed for bankruptcy within 3 years. The number of positive samples is 495 (4.71% of total), and the total number of missing cells is 9,888. The data dictionary file (data_dictionary.csv) provides the financial definition, category, and missing rate for each ratio. Notes: The data is based on Polish filings, with bankrupt companies covering 2000-2012 and surviving companies covering 2007-2013. The absolute default rate is only applicable to this group and should not be regarded as current rates for other markets. This dataset is suitable for tabular classification tasks such as bankruptcy prediction and credit risk assessment.
数据集概述
基本信息
- 数据集名称: weiqi001/bankruptcy-early-warning-data
- 显示名称: Polish Companies Bankruptcy (3-year horizon)
- 许可证: CC-BY-4.0
- 任务类别: 表格分类(tabular-classification)
- 标签: 金融(finance)、信用风险(credit-risk)、破产预测(bankruptcy-prediction)
- 数据规模: 10K<n<100K
数据内容
- 来源: UCI机器学习库数据集365(Polish Companies Bankruptcy Data),由
src/prepare_data.py处理生成 - 行数: 10,503个公司-年份样本
- 列数: 65列(64个财务比率 + 1个目标变量)
- 目标变量:
bankrupt(表示公司在3年内申请破产) - 正例数量: 495(占比4.71%)
- 缺失单元格: 9,888个
- 文件校验: SHA-256值为
16aee66f41ae1417e0bbcee72ac2e5521736dfe98463888fff6e9f5345b048dc
数据处理说明
- 数据来源于
3year.arffARFF文件,已解析为CSV格式 - 删除了没有可用财务比率的行
- 没有进行缺失值填充或数据变换,属于原始分析表
- 数据字典文件
data_dictionary.csv提供了每个财务比率的定义、所属类别和缺失率
注意事项
- 数据基于波兰公司 filings,破产企业为2000-2012年,存活企业为2007-2013年
- 绝对违约率是针对该人群校准的,不应视为其他市场或当前市场的实际违约率
引用信息
Zieba, M., Tomczak, S. K., & Tomczak, J. M. (2016). Ensemble boosted trees with synthetic features generation in application to bankruptcy prediction. Expert Systems with Applications, 58, 93-101.



