Journal
KNOWLEDGE-BASED SYSTEMS
Volume 193, Issue -, Pages -Publisher
ELSEVIER
DOI: 10.1016/j.knosys.2019.105454
Keywords
Dense members; Local cores; Density peaks; Clustering
Categories
Funding
- Project of Chongqing Natural Science Foundation, China [cstc2019jcyj-msxmX0683]
- National Natural Science Foundation of China [61601060, 61702060]
- Project of Chongqing Municipal Education Commission, China [KJZH17104, KJZD-M20190140]
Ask authors/readers for more resources
An efficient clustering algorithm by fast search and find of density peaks (DP) was proposed and attracted much attention from researchers. It assumes that cluster centers are surrounded by lower density points and have a larger distance from points with higher densities. According to the characteristic of cluster centers, we can easily obtain centers from decision graph. However, DP algorithm fails to cluster manifold data sets, especially when there are a lot of noises in the manifold data sets. In this paper, we propose a dense members of local core-based density peaks clustering algorithm DLORE-DP. First, we find local cores to represent the data set. After that, only dense members of local cores are taken into consideration when computing the graph distance between local cores, avoiding the interference of noises. Then, natural neighbor-based density and the new defined graph distance are used to construct decision graph on local cores and DP algorithm is employed to cluster local cores. Finally, we assign each remaining point to the cluster its representative belongs to. The new defined graph distance helps our algorithm cluster manifold data sets and the elimination of low density points makes it more robust. Moreover, since we only calculate the graph distance between local cores, instead of all pairs of points, it greatly reduces the running time. The experimental results on synthetic and real data sets show that DLORE-DP is more effective, efficient and robust than other algorithms when clustering manifold data sets with noises. (c) 2020 Elsevier B.V. All rights reserved.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available