The Experimental Design of Classification Models: An Application of Recursive Partitioning and Bootstrapping to Commercial Bank Loan Classifications
探讨分类模型实验设计中的三个关键要素：分类错误的损失函数、分类算法和误分类损失估计方法，并以商业银行贷款分类为例说明其重要性。
This paper examines several issues in the experimental design and empirical testing of classification models. As an illustration, we focus on the classification of commercial bank loans. We stress the importance of and the interactions among three elements: the loss function associated with classification errors, the algorithm used to discriminate among or predict classifications, and the method used to estimate the expected misclassification losses achieved by various algorithms. These aspects of the design are of particular importance when an empirical researcher must proceed without the guidance of a well-specified model of the phenomenon under investigation. We use two nonparametric, computer-intensive statistical techniques