SUDEPHIC: Self-tuning density-based partitioning and hierarchical clustering

Abstract

Clustering is one of the primary techniques in data mining, for which to find the user expecting result is a major issue. However, to dynamically specify the parameters for clustering algorithms presents an obstacle for users. This paper firstly introduces a novel density-based partitioning and hierarchical algorithm, which makes it easy to employ synthetic feedback mechanism in clustering. Additionally, by investigating into the relation between parameters and the clustering result, we propose a self-tuning technique for the setting of parameters. Meanwhile, the density distribution within a cluster can be expressed in the result for the user to specify the cluster's feature. The algorithm is both evaluated in theory and practice. It outperforms many existing algorithms both in efficiency and quality. © Springer-Verlag 2004.