A novel clustering method based on spatial operations

Abstract

In this paper we present a novel clustering method that can deal with both numerical and categorical data with a novel clustering objective and without the need of a user specified parameter. Our approach is based on an extension of database relation - hyperrelations. A hyperrelation is a set of hypertuples, which are vectors of sets. In this paper we show that hyperrelations can be exploited to develop a new method for clustering both numerical and categorical data. This method merges hypertuples pairwise in the direction of increasing the density of hypertuples. This process is fully automatic in the sense that no parameter is needed from users. Initial experiments with artificial and real-world data showed this novel approach is promising. © Springer-Verlag Berlin Heidelberg 2006.