A Possibilistic Density Based Clustering for Discovering Clusters of Arbitrary Shapes and Densities in High Dimensional Data

Abstract

Apart from the interesting problem of finding arbitrary shaped clusters of different densities, some applications further introduce the challenge of finding overlapping clusters in the presence of outliers. Fuzzy and possibilistic clustering approaches have therefore been developed to handle such problem, where possibilistic clustering is able to handle the presence of outliers compared to its fuzzy counterpart. However, current known fuzzy and possibilistic algorithms are still inefficient to use for finding the natural cluster structure. In this work, a novel possibilistic density based clustering approach is introduced, to identify the degrees of typicality of patterns to clusters of arbitrary shapes and densities. Experimental results illustrate the efficiency of the proposed approach compared to related algorithms.