A novel multi-stage ensemble model with voting-based outlier adaptation and balanced sampling for credit scoring

Three datasets from the UC Irvine (UCI) machine learning repository, that is, the Australian, German, and Japanese datasets, were adopted for the current study. The Australian credit dataset contains 690 samples, of which 307 are positive and 383 are negative. The dimensions of its input features are 15. The German credit dataset contains 1000 samples, 700 of which are positive and 300 are negative. The dimensions of its input features are 21. The Japanese credit dataset contains 690 samples, of which 383 are positive and 307 are negative. The dimensions of its input features are 16.