Fuzzy Rule-Based Classification with Hypersphere Information Granules

Abstract

Fuzzy rule-based classification has been studied by a number of classification architectures. In this study, hypersphere information granules are used to form initial fuzzy classification model in an intuitive and interpretative way. The principle of justifiable granularity offers a certain way to optimizing information granules while facing the coverage and specificity criteria. By engaging a synergy of the principle of justifiable granularity and migrating prototypes, the refined classification model is constructed for better classification performance. A series of experiments concerning synthetic datasets and comparative studies are also implemented to exhibit the feasibility and effectiveness of the proposed classification method.