A Hierarchical Classification Model Based on Granular Computing

Abstract

In this paper, after a brief overview of the existing methods, we present a new hierarchical classification algorithm based on quotient space theory of the granular computing. This algorithm deals with the samples from coarse to fine both in the training and testing processes. A group of classifiers are firstly trained by the samples generated under different quotient space. Then the trained classifiers will be used to label the testing samples set hierarchically. In our method, Support Vector Machines is chosen to acquire the discrimination function between two classes in the training processes. And the hypercubes which represent support vectors are subdivided to generate the samples set for training and testing under different quotient space. Finally, experimental results have substantiated the effectiveness of the proposed method.