Aggregated center-based clustering algorithm based on principal component radius

Mingchang Cheng

doi:10.1016/j.neucom.2025.129469

Aggregated center-based clustering algorithm based on principal component radius

2025-04-01

0

PRE

AI

DOI：10.1016/j.neucom.2025.129469

Original

Original request for help

Save

Abstract

En 中文

Center-based clustering algorithms such as k-means have strict requirements on data distribution. As an important initialization improvement method, k-means++ can effectively overcome the problem of local convergence, but it still need to accurately specify the number of clusters and not applicable to non-convex clusters. In order to overcome the above limitations, this paper proposes an aggregated center-based clustering algorithm based on the principle component radius (PCR). First, k-means++ is used as the initialization method to generate an excess of clusters. Subsequently, a cluster characterization method, PCR, with shrinking effect is proposed to accurately characterize the core region of clusters, guiding cluster merging. On this foundation, a graph based aggregation process can efficiently complete cluster identification and adaptively determine the number of clusters. Experimental results on all synthetic and real-world datasets verify the superiority of the proposed algorithm and its high tolerance to parameters and initialization.

Keywords:

Clustering

Principal component radius

Aggregation process

k-means plus plus

Aggregated center-based clustering algorithm based on principal component radius

Abstract

Journal

Organization