Evaluating the Ambiguities of Class Structure via Euclidean Distance

Abstract

The classification is a supervised learning approach in the machine learning and, therefore, the class structure was specified by the domain experts manually in advance. The goal of this paper is to evaluate the degree of ambiguity between any two classes in the existing class structure while the similarity between two classes was estimated via Euclidean distance. In this paper, Distinguishable Distance Ratios (DDR) and Class Ambiguity Ratio (CAR) between any two classes are proposed to indicate the degree of the ambiguity between classes. The degree of class ambiguity between two classes supposed to be high if the value of DDR is low and the value of CAR is high. The experimental resources for class structure evaluation includes ”Iris Plant”, ”Wine Recognition” and ”Glass Identification”, and the DDR and CAR did reveal the degree of class ambiguity. This works offer domain expertise an approach to examine the fitness of class structure if necessary.