A Hybrid Approach To Hierarchical Density-based Cluster Selection

11/06/2019

∙

by   Claudia Malzer, et al.

∙

0

∙

share

HDBSCAN is a density-based clustering algorithm that constructs a cluster hierarchy tree and then uses a specific stability measure to extract flat clusters from the tree. We propose an alternative method for selecting clusters from the HDBSCAN hierarchy. Our approach, HDBSCAN(ϵ̂), is particularly useful for data sets with variable densities where we require a low minimum cluster size but want to avoid an abundance of micro-clusters in high-density regions. The method uses an additional input parameter ϵ̂ and acts like a hybrid between DBSCAN* and HDBSCAN. It can easily be integrated into existing HDBSCAN implementations.