A novel structural mass based dissimilarity measure

作者：Fang Peng; Huang Liusheng; Xu Hongli; Wang Shaowei

来源：5th International Conference on Computer Science and Network Technology, ICCSNT 2016, 2016-12-10 To 2016-12-11.

DOI：10.1109/ICCSNT.2016.8070169

• clustering
• Data dependent
• DBSCAN
• dissimilarity
• Manifold structures

Data dependent dissimilarity provides a better closest adaptation than distance measures. When dealing with arbitrary types of data sets especially those with manifold structures, mass-based dissimilarity [1] cannot perform well. Taking the structure into account, this paper introduces a generic structural mass-based dissimilarity which is easily applied to existing algorithms in different missions. It increases the similarities between instances within the same structure and reduces the similarities between instances in different structures. Compared with mass-based dissimilarity, structural mass-based dissimilarity improves the accuracy rate when applied to various clustering methods. Our experiment results demonstrate that the proposed method is efficient and robust for more applications and outperforms existing approaches.

• 出版日期2016
• 单位中国科学技术大学