An algorithm to adaptive determination of density threshold for density-based clustering

Density-based clustering is a good solution for classifying enormous data without any early knowledge. In this paper, the focus is distance measure of dense regions, which can be set adaptively. By translating a density threshold selection problem into a determination problem of the spherality radius, a novel density-based clustering algorithm based on determination of density threshold is proposed. First, we propose the definition of the partial cluster which is a radius threshold composed of a single object. Second, based on the analysis of radius threshold range from dense transitive closure via Hopkins statistic techniques, dense regions are formed by several partial cluster. Then we realize clustering according to the fact whether these regions can be recognized as clusters. Finally, the clustering results on various datasets are presented and its future direction is also discussed.