Improving the Cluster Structure Extracted from OPTICS Plots

Density-based clustering is closely associated with the two algorithms DBSCAN and OPTICS. While the first finds clusters connected at a single density threshold, the latter allows the extraction of a cluster hierarchy based on different densities. Extraction methods for clusters from OPTICS rely on an intermediate representation, known as the OPTICS plot. In this plot, which can be seen as a density profile of the data set, valleys (areas of higher density) are associated with clusters. Multiple methods for automatic detecting such valleys have been proposed, but all of them tend to produce a particular artifact, where some point is included in the cluster that may be far away from the remainder. In this article, we will discuss this commonly seen artifact, and propose a simple but sound way of removing the artifacts. At the same time, this change is minimally invasive, and tries to keep the existing algorithms largely intact for future study.