A novel procedure to identify the minimized overlap boundary of two groups by DEA model

Abstract

In discriminant analysis (DA), the overlap among groups is a major source of misclassification. The identification of overlap boundary may provide additional information for the decision makers in risk management. Most effort of previous researches aim to improve the hit-ratio of correct classification, but few concentrates on the identification of overlap boundary. In this paper, a novel procedure which is based on data envelopment analysis (DEA) is proposed to identify the overlap boundary of two groups. The minimized overlap boundary can be obtained after taking linear transformation. An important merit of the proposed approach is to resolve the problem of calibrating overlap boundary with parametric approach in the case of small sample size. © Springer-Verlag Berlin Heidelberg 2005.