Comparing metrics in fuzzy clustering for symbolic data on SODAS format

Abstract

A number of approaches to solve the problem of data clustering are available in the literature. This paper introduces a comparative study in some distances metrics very known in the literature of symbolic data. This work uses an adaptation of an algorithm that applies concepts of fuzzy clustering and then creates groups of individuals characterized by symbolic variables of mixed types. The core of the algorithm consists of a dissimilarity function that can be replaced without collateral effects. The input data is provided on the format of a SODAS file. The results of the experiments on representative databases show the performance of each analyzed metric. © Springer-Verlag Berlin Heidelberg 2004.