Full Text: PDF
DOI: 10.23952/cot.2021.15
Received October 26, 2021; Accepted November 9, 2021; Published November 22, 2021
Abstract. The K-means clustering algorithm as the representation of the partition algorithms is efficient for convex data. Density-based clustering algorithms can be used to solve the nonconvex data, but they adopt the global threshold value, which makes them perform badly on varying densities. In particular, the densities of components are very different in datasets. Aiming at these problems, we propose a Density-Based Core-Structures Expansion algorithm (DBCSE) and use the Density-Based Core-Structures structured by density searching to replace the represent points of K-means. And we set a relatively large density threshold to detect the core structures and adopt border expanding way to cluster relatively low density area so that our algorithm can cluster different density components, arbitrary shape data, and detect noise. We also conduct experiments on two synthetic datasets and four UCI datasets to demonstrate our algorithm effectiveness.
How to Cite this Article:
Lei Chen, Density-based core-structures expansion for clustering data with varying densities greatly, Communications in Optimization Theory, Vol. 2021 (2021), Article ID 15, pp. 1-12.