A Novel Evaluation Criterion for Density Clustering via Circular Information Granules
Sheng Du, Cheng Huang, Zixin Huang, Witold Pedrycz · IEEE Signal Processing Letters · 2025
Density clustering is a pivotal algorithm for data clustering and analysis, finding extensive and significant industrial application. There are two key adjustable parameters in density clustering: cluster radius and minimum number of cluster points. At present, the selection of more suitable parameters predominantly depends on statistical methods and analysis, which lacks a precise and effective evaluation criterion. In this paper, a novel evaluation criterion for density clustering via circular information granules is proposed. It constructs circular information granules based on the density clustering results through the principle of justifiable granularity, and then finds the largest sum of volumes of circular information granules. Consequently, it determines the optimal clustering radius and the minimum number of clustering points. Experimental results show that the proposed method provides a more comprehensive evaluation of density clustering results compared to the existing evaluation criterion.