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.
| Original language | English |
|---|---|
| Pages (from-to) | 577-586 |
| Number of pages | 10 |
| Journal | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
| Volume | 3483 |
| Issue number | IV |
| DOIs | |
| State | Published - 2005 |
| Event | International Conference on Computational Science and Its Applications - ICCSA 2005 - , Singapore Duration: 9 May 2005 → 12 May 2005 |