UIFDBC: Effective density based clustering to find clusters of arbitrary shapes without user input

Density-based clustering has the ability to detect arbitrary shaped clusters in any dataset. In recent years, several density peak clustering methods have been reported. Among these, a few need user input(s), but majority use cluster validity indices to provide the best results. In this paper, we propose a density-based user-input-free clustering method named UIFDBC, which is capable of detecting clusters of arbitrary shapes, without depending on any specific cluster validity index. The method is evaluated on 16 synthetic and 7 real-world datasets and compared with 8 recent density-based clustering methods. The results show our method is superior, in general, to its counterparts in terms of discovering arbitrary shaped clusters on tested datasets. The approach also has the ability to handle low-density instances in a special manner to minimize error propagation. Our method is available as an R package and can be downloaded by clicking the link https://sites.google.com/view/hussinchowdhury/software.

Automated summary

The study introduces a density-based clustering method UIFDBC that can detect clusters of arbitrary shapes without user input. Evaluation results show the method outperforms its counterparts in discovering arbitrary shaped clusters and has the ability to handle low-density instances.