Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
Spherical Spectral Clustering Based on Density and Neighbor Rule
School of Computer Science and Engineering, Northwest Normal University, China.
ABSTRACT
Spectral clustering is a valuable technique for data with complex structures. However, it still faces limitations, such as generating overly large similarity matrices for large-scale datasets and exhibiting poor performance on datasets with uneven density distributions. To address these issues, we propose a spherical spectral clustering algorithm based on density and neighbor rule (SSCNN). This algorithm divides the data into spheres by calculating the initial cluster compactness, and the similarity between the data is replaced by the similarity between the balls, thus greatly reducing the similarity matrix. Meanwhile, the similarity is calculated according to the defined ball density and shared neighbor rule, and the adjacent balls with similar density are merged, so as to solve the wrong classification caused by close distance but large density difference. Finally, experiments are carried out on Synthetic datasets and UCI datasets, which exhibit that SSCNN has good adaptability to large-scale datasets with large density variations.
Keywords: Spectral clustering, Multi-granularity ball, Density, Ball-Kmeans, Mutual nearest neighbors, Shared neighbor.
Download PDF
School of Computer Science and Engineering, Northwest Normal University, China.
ABSTRACT
Spectral clustering is a valuable technique for data with complex structures. However, it still faces limitations, such as generating overly large similarity matrices for large-scale datasets and exhibiting poor performance on datasets with uneven density distributions. To address these issues, we propose a spherical spectral clustering algorithm based on density and neighbor rule (SSCNN). This algorithm divides the data into spheres by calculating the initial cluster compactness, and the similarity between the data is replaced by the similarity between the balls, thus greatly reducing the similarity matrix. Meanwhile, the similarity is calculated according to the defined ball density and shared neighbor rule, and the adjacent balls with similar density are merged, so as to solve the wrong classification caused by close distance but large density difference. Finally, experiments are carried out on Synthetic datasets and UCI datasets, which exhibit that SSCNN has good adaptability to large-scale datasets with large density variations.
Keywords: Spectral clustering, Multi-granularity ball, Density, Ball-Kmeans, Mutual nearest neighbors, Shared neighbor.
Download PDF