A novel multi-stage ensemble model with enhanced outlier adaptation for credit scoring

Nine datasets from the UC Irvine (UCI) machine learning repository, i.e., the Australian, Japanese, German (Asuncion & Newman, 2007), Taiwan (Yeh & Lien, 2009) and Polish credit datasets (Zięba et al., 2016) were adopted for the current study. The Polish credit datasets contain five datasets distinguished five classification cases that depend on the forecasting period (e.g., the Polish 1, the Polish 2, the Polish 3, the Polish 4 and the Polish 5). AER credit dataset (Greene, 2003), which is a credit card dataset for econometric analysis. Creator dataset, which is published in 2019 by a Chinese digital government services provider named Creator Information Technology Co., Ltd[1]. The Creator dataset contains the property rights, financial statements, and basic company information of 35960 Chinese companies.

[1] http://www.chinacreator.com/cn/