A novel path-based clustering algorithm using multi-dimensional scaling

Abstract

Data clustering is a difficult and challenging task, especially when the hidden clusters are of different shapes and non-linearly separable in the input space. This paper addresses this problem by proposing a new method that combines a path-based dissimilarity measure and multi-dimensional scaling to effectively identify these complex separable structures. We show that our algorithm is able to identify clearly separable clusters of any shape or structure. Thus showing that our algorithm produces model clusters; that follow the definition of a cluster. © Springer-Verlag Berlin Heidelberg 2009.