Extended fast search clustering algorithm: widely density clusters, no density peaks

TL;DR

This paper introduces E_CFSFDP, an extension of the CFSFDP clustering algorithm, which effectively handles datasets with multiple density peaks per cluster by combining initial clustering with hierarchical merging.

Contribution

The paper proposes a novel extension to CFSFDP that improves clustering performance on complex datasets with no single density peak, inspired by hierarchical clustering methods.

Findings

E_CFSFDP outperforms original CFSFDP on datasets with multiple density peaks.

The approach effectively merges sub-clusters to handle 'no density peaks' scenarios.

Experimental results demonstrate improved clustering accuracy on challenging datasets.

Abstract

CFSFDP (clustering by fast search and find of density peaks) is recently developed density-based clustering algorithm. Compared to DBSCAN, it needs less parameters and is computationally cheap for its non-iteration. Alex. at al have demonstrated its power by many applications. However, CFSFDP performs not well when there are more than one density peak for one cluster, what we name as "no density peaks". In this paper, inspired by the idea of a hierarchical clustering algorithm CHAMELEON, we propose an extension of CFSFDP,E_CFSFDP, to adapt more applications. In particular, we take use of original CFSFDP to generating initial clusters first, then merge the sub clusters in the second phase. We have conducted the algorithm to several data sets, of which, there are "no density peaks". Experiment results show that our approach outperforms the original one due to it breaks through the strict claim of data sets.