Density Peaks Clustering Algorithm Based on Neighborhood Radius and Membership Degree

Abstract

The density peaks clustering (DPC) algorithm is a density-based clustering method that effectively identifies clusters with uniform densities. However, if the datasets have uneven density, clusters with lower densities tend to have lower decision values, which often leads to the cluster centers being overlooked. To address this limitation, a novel density peaks clustering algorithm incorporating neighborhood radius and membership degree is proposed. The method begins by introducing k-nearest neighbor density estimation to establish a density threshold, segmenting datasets into high-density and low-density regions. In the high-density region, the DPC algorithm is applied to perform initial clustering, identifying prominent cluster structures. For low-density points, neighborhood radius and density criteria are employed to assign these points to appropriate high-density clusters, thereby reducing misclassification. In addition, the membership degree concept is incorporated to improve the accuracy of low-density point assignments. Low-density points that remain unassigned undergo secondary clustering using the DPC algorithm. The proposed approach is evaluated on eight synthetic datasets and eleven real-world datasets, with comparisons to DPC-KNN, DPC, K-means, and DBSCAN. The experimental results demonstrate that the proposed algorithm consistently outperforms these methods in clustering performance, highlighting its effectiveness and robustness.