Optimum Feature Selection for Decision Functions.

Feature selection is one of the most important processes in the design of pattern classifiers. This paper presents an optimum feature selection method which is applicable to arbitrary (nonlinear) decision functions. It is assumed that a finite number of trains lag samples (training set) is given for each pattern class, and the decision function is designed based on the training sets. The training sets are edited by renoving the samples which are classified incorrectly by the decision function. That the feature selection problem is transformed to a modified zero-one integer program. In this method, under a chosen permissible error, a minimum feature subset can be found which is combinationally optimum. Numerical, examples of feature selection I or a linear and a quadratic decision function are presented. [ABSTRACT FROM AUTHOR]