CA-CSM: a novel clustering algorithm based on cluster center selection model

Clustering by fast search and find of density peaks (DPC) is a well-known algorithm due to the simple structure and high extensibility. It requires neither iteration nor additional parameters. However, DPC and most of its improvements still encounter some challenges such as parameter dependence, unreasonable metric and cluster center determination difficulty. Aiming at these issues, we propose a novel clustering algorithm based on cluster center selection model (CA-CSM). Firstly, we calculate the density parameter automatically according to the local information of the object to reduce the parameter dependence. Subsequently, we propose the concept of boundary degree to discriminate core objects from non-core objects. With the local density metric, we establish a model (CSM) with high expansibility to automatically detect the cluster centers from core objects. We test CA-CSM on 21 datasets using five benchmarks and compare it to 7 state-of-the-art algorithms. Extensive experiments and analysis show that our algorithm is feasible and effective.