On performance of case-based reasoning in Chinese business failure prediction from sensitivity,specificity, positive and negative values

Case-based reasoning (CBR) is a machine learning technique of high performance in classification problems, and it is also a chief method in predicting business failure. Recently, several techniques have been introduced into the life-cycle of CBR for business failure prediction (BFP). The drawback of former researches on CBR-based BFP is that they only use total predictive accuracy when assessing CBR. In this research, we provide evidence on performance of CBR in Chinese BFP from various views of sensitivity, specificity, positive and negative values. Data are collected from Shanghai Stock Exchange and Shenzhen Stock Exchange in China. And we present how data are preprocessed from the view of data mining. The classical CBR model on the base of Euclidean metric, the grey CBR model on the base of grey coefficient metric, and the pseudo CBR model on the base of pseudo outranking relations are employed to make a comparative study on CBR's predictive performance in BFP. Meanwhile, support vector machine (SVM) is employed to be a baseline model for comparison. The results indicate that pseudo CBR produces better performance in Chinese BFP than classical CBR and grey CBR significantly on the whole, and it outperforms SVM marginally by total predictive accuracy and sensitivity, while it is not significantly worse than SVM by specificity.