Improved Linear Programming Formulations for the Multi-group Discriminant Problem

Abstract

This paper sets out to solve the multi (more than two)-group classification problem, and develops a new linear programming model which simultaneously determines the cut-off values for the different classification functions. Instead of decomposing the content in the multi-group problem to facilitate computation of the cut-off values, this new model aggregates information contained in the multi-group problem which, intuitively, should provide better estimates of the group boundaries. Furthermore, this new model, one existing LP model, and a statistical approach will be tested by using real-life data.