Discovery of Arbitrary-Shapes Clusters Using DENCLUE Algorithm

Abstract

One of the main requirements in clustering spatial datasets is the discovery of clusters with arbitrary-shapes. Density-based algorithms satisfy this requirement by forming clusters as dense regions in the space that are separated by sparser regions. DENCLUE is a density-based algorithm that generates a compact mathematical form of arbitrary-shapes clusters. Although DENCLUE has proved its efficiency, it cannot handle large datasets since it requires large computation complexity. Several attempts were proposed to improve the performance of DENCLUE algorithm, including DENCLUE 2. In this study, an empirical evaluation is conducted to highlight the differences between the first DENCLUE variant which uses the Hill-Climbing search method and DENCLUE 2 variant, which uses the fast Hill-Climbing method. The study aims to provide a base for further enhancements on both algorithms. The evaluation results indicate that DENCLUE 2 is faster than DENCLUE 1. However, the first DECNLUE variant outperforms the second variant in discovering arbitrary-shapes clusters