Two step semi-optimal branch and bound algorithm for feature selection in mixed variable discrimination

摘要

The paper presents a two step procedure for the selection of the most discriminative discrete as well as continuous features in the location model for mixed-variable discriminant analysis. The multivariate discriminatory measure T2 is used as a criterion for subset choice.The first step consists in finding stepwisely the most discriminative discrete features. The second step enables one to select the subset of continuous variables for the previously chosen discrete feature set. The two steps may be reversed. When the number of discrete variables is already fixed the selection of continuous features is made efficiently by the branch and bound algorithm.