Proposal of Class Determination Method for Generated Virtual Data in pdi-Bagging

Abstract

We have already proposed pdi-Bagging as one of ensemble learning methods of clustering. The accuracy of the classifier strongly depends on the generation type, generation position and generation class of virtual data. However, the accuracy is not stable in the correct virtual data type because the virtual data generate in the wide area of the data space. Also, the accuracy of the evaluation index is not high because the evaluation is defined in each dimension. In this paper, we propose a new method to specify the generation area of virtual data and change the generation class of virtual data. Specifically, the error type virtual data generates in the area of the discrimination line. The occurrence area is specified near the distribution center of the correct discrimination virtual data. In addition, we define a new evaluation formula that introduces the degree of similarity with the correct - error virtual data in multidimensional space. We formulate a new pdi-Bagging algorithm that introduces these methods, and discuss the usefulness of this method using numerical examples.