Document Type : Research Note

Authors

• Mostafa Ghazizadeh-Ahsaee
• Afsaneh Shamsaddini-Farsangi

Department of Computer Engineering, Shahid Bahonar University of Kerman, Iran

Abstract

Clustering datasets with varying densities is challenging because classical Density Peak Clustering (DPC) may fail to identify valid centers in sparse regions and may incorrectly assign sparse samples to dense clusters. To address this issue, this paper proposes a Density Peak Clustering algorithm based on Fuzzy Shared Nearest Neighbors (DPC-FSNN). The proposed method combines k-nearest neighbors and fuzzy neighborhood information to construct a fuzzy nearest neighbor kernel for adaptive local density estimation. It also incorporates shared nearest neighbor information into the local distance computation and sample assignment process to improve center detection and assignment reliability. Experiments on varying-density datasets (e.g., Jain, Compound, and LineBlobs), complex-shape datasets, and real-world datasets are evaluated using AMI, ARI, and FMI. The results show that DPC-FSNN achieves competitive and often superior clustering performance compared with IDPC-FA, FKNN-DPC, DPC-FWSN, DPCSA, FNDPC, and the classical DPC algorithm. A computational complexity analysis is also provided, showing that the proposed method has overall complexity O(n^2) in the standard implementation.

Keywords

• Clustering
• Multi-density
• Nearest Neighbor
• Fuzzy Neighborhood

Main Subjects

• H.6.4. Clustering