A Bi-criteria Dimension Reduction Approach with Application in Supplier Selection

Abstract

The supplier selection problem usually involves many criteria. It is desired to exclude some redundant criteria so as to simplify decision analysis without losing too much useful information. This process is called the dimension reduction. The principal component analysis is a typical dimension reduction method but it does not really reduce the dimension of the problem since each principal component is actually a linear combination of the original indicators. This paper presents an approach to substantially reduce the number of criteria. The approach involves two criteria: correlation between criteria and variability in criteria scores. The former reflects information redundancy and the latter reflects the discriminative capability of a criterion. A threshold model is developed for determining whether a criterion should be included for further analysis or not. The approach is illustrated by an example.