Improved Data Classification using Fuzzy Euclidean Hyperbox Classifier

Abstract

In this paper, a modification to the simple fuzzy min-max classifier has been proposed. The primary objective was the design of an efficient classifier that can be used in a wide range of application domains, unlike most prior works which focus on selected problems. This work retains the fuzzy neural structure of the original work but proposes a different membership function for the hyperboxes based on the Euclidean distance measure. The new function takes into consideration the centroids of the hyperboxes and not just the min and max points. The competence of the proposed classifier is tested on kinds of datasets. Further, a novel approach in which the classifier can also handle partly labeled data (or data with missing labels) is also discussed. One of the most important requisites of any classification algorithm is its efficiency. In the result-driven technological world of today, where mobile computing is a major thrust area, simple and elegant solutions are highly sought. Thus speed and efficiency were major considerations in the choice of the classifier for the classification system designed.