Tehnički vjesnik, Vol. 25 No. 2, 2018.
Izvorni znanstveni članak
https://doi.org/10.17559/TV-20171121052315
Improved Density Peak Clustering Algorithm Based on Choosing Strategy Automatically for Cut-off Distance and Cluster Centre
Limin Wang
; School of Management Science and Information Engineering, Jilin University of Finance and Economics, Jilin Province Key Laboratory of Fintech, Changchun 130117, China
Mingyang Li
; School of Management Science and Information Engineering, Jilin University of Finance and Economics, Jilin Province Key Laboratory of Fintech, Changchun 130117, China
Xuming Han
; School of Computer Science and Engineering, Changchun University of Technology, Changchun 130012, China
Ruihong Zhou
; School of Management Science and Information Engineering, Jilin University of Finance and Economics, Jilin Province Key Laboratory of Fintech, Changchun 130117, China
Kaiyue Zheng
; School of Management Science and Information Engineering, Jilin University of Finance and Economics, Jilin Province Key Laboratory of Fintech, Changchun 130117, China
Meihan Liu
; School of Management Science and Information Engineering, Jilin University of Finance and Economics, Jilin Province Key Laboratory of Fintech, Changchun 130117, China
Sažetak
Due to the defect of quick search density peak clustering algorithm required an artificial attempt to determine the cut-off distance and circle the clustering centres, density peak clustering algorithm based on choosing strategy automatically for cut-off distance and cluster center (CSA-DP) is proposed. The algorithm introduces the improved idea of determining cut-off distance and clustering centres, according to the approximate distance that maximum density sample point to minimum density sample point and the variation of similarity between the points which may be clustering centres. First, obtaining the sample point density according to the k-nearest neighbour samples and tapping the sample sorting of the distance to the maximum density point; then finding the turning position of density trends and determining the cut-off distance on the basis of the turning position; finally, in view of the density peak clustering algorithm, finding the data points which may be the centres of the cluster, comparing the similarity between them and determining the final clustering centres. The simulation results show that the improved algorithm proposed in this paper can automatically determine the cut-off distance, circle the centres, and make the clustering results become more accurate. In the end, this paper makes an empirical analysis on the stock of 147 bio pharmaceutical listed companies by using the improved algorithm, which provides a reliable basis for the classification and evaluation of listed companies. It has a wide range of applicability.
Ključne riječi
clustering center; cut-off distance; Density Peak Clustering Algorithm; maximum density; similarity
Hrčak ID:
199153
URI
Datum izdavanja:
21.4.2018.
Posjeta: 1.516 *