An optimization model for constrained discriminant analysis and numerical experiments with iris, thyroid, and heart disease datasets.

Abstract

A nonlinear 0/1 mixed integer programming model is presented for a constrained discriminant analysis problem. The model enables controlling misclassification probabilities by placing restrictions on the numbers of misclassifications allowed among the training entities and incorporating a "reserved-judgment" region to which entities whose classifications are difficult to determine may be allocated. A linearization of the model is given, and preliminary numerical results for two medical and one non medical domain are presented.