| Summary: | 博士 === 輔仁大學 === 商學研究所 === 96 === Analyzing high-dimensional bank data to discover valuable information has long been recognized as a very difficult and challenging task. Accordingly, this study attempts to propose a three-stage ensemble classification method which incorporates multivariate adaptive regression splines(MARS) serving as a vehicle of feature selection, nonparametric weighted feature extraction(NWFE) for dimension reduction, and support vector machines(SVMs) as a classifier in constructing a cardholder behavioral scoring model. The major purpose of doing so is discerning customers’ future repayment status with desired classification accuracy, low misclassification costs and shortened model computation time. Analytical results reveal that proposed three-stage classification model outperforms the conventional and innovative discriminant analysis, artificial neural networks, MARS and SVMs techniques under various performance criteria. In addition, it is also noted that the proposed method can also provide managerial implications for real world practices.
|
|---|
