weiqi001/bankruptcy-early-warning-data
weiqi001/bankruptcy-early-warning-data The processed table behind the Corporate Bankruptcy Early-Warning System. Derived from UCI ML Repository dataset 365 by src/prepare_data.py: the 3year.arff ARFF file parsed to CSV, with rows carrying no usable ratios dropped. No values are imputed or transformed here - that happens inside the model pipeline, so this file is the raw analysis table. Rows 10,503 firm-years Columns 65 (64 ratios + target) Target bankrupt -… See the full description on the dataset page: https://huggingface.co/datasets/weiqi001/bankruptcy-early-warning-data.
weiqi001/bankruptcy-early-warning-data
The processed table behind the Corporate Bankruptcy Early-Warning System.
Derived from UCI ML Repository dataset 365 by src/prepare_data.py: the 3year.arff ARFF file parsed to CSV, with rows carrying no usable ratios dropped. No values are imputed or transformed here - that happens inside the model pipeline, so this file is the raw analysis table.
data_dictionary.csv gives the financial definition, family and missing rate of every ratio.
Caveat
Polish filings, bankrupt firms 2000-2012 and survivors 2007-2013. Absolute default rates are calibrated to that population and should not be read as current rates for another market.
Citation
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.