Density peak clustering with attribute-level similarity and two-stage allocation

Clustering by fast search and find of density peaks (DPC) is an unsupervised clustering method. This algorithm is based on the assumption that cluster centers have high local densities and are generally far from each other. Cluster centers can be easily located and clustering can be achieved using a decision graph. While effective, it suffers from two key limitations: its reliance on Euclidean distance fails to capture attribute-level heterogeneity, and the single-pass allocation is prone to error propagation. To address these problems, a novel density peak clustering with attribute-level similarity and two-stage allocation (CKDPC) algorithm is proposed in this paper. Firstly, a new attribute-level similarity measure is introduced to redefine local density, enhancing sensitivity to heterogeneous feature contributions. Secondly, a two-stage allocation strategy is designed that integrates attribute-based density similarity and distance information, thereby reducing boundary misassignments. Finally, we benchmark the proposed clustering algorithm against state-of-the-art DPC variants and classical methods on standard synthetic and real-world datasets. The experimental results demonstrate that our proposed clustering algorithm can find cluster centers, recognize clusters regardless of their shape and dimension of the space in which they are embedded, be unaffected by outliers, and can often outperform benchmark algorithms, verifying the efficiency and superiority of the proposed CKDPC.