Gradient method for determining non-informative features on the basis of a homogeneous criterion with a positive degree

Abstract

Abstract The choice of informative features is one of the main tasks in the problem of pattern recognition and the effectiveness of its solution in many respects depends on the criterion of information content used. Some methods for choosing informative features are implemented by eliminating non-informative features from the feature space. A single criterion has not been developed to select a set of non-informative features. Therefore, the criterion is selected from the formulation of a practical problem. Currently, several methods have been developed for selecting features that are focused on the use of a specific information content criterion. To date, more than 50 homogeneous criteria with zero order have been developed, but in many practical tasks the task of choosing non-informative features requires the use of homogeneous criteria with a positive degree. The paper proposes a gradient method and an algorithm for determining non-informative features based on a homogeneous criterion with a positive degree.