A Novel Self-Adaptive Affinity Propagation Clustering Algorithm Based on Density Peak Theory and Weighted Similarity

To solve both the similarity calculation method and parameter limits problems of the affinity propagation algorithm (AP), the self-adaptive affinity propagation clustering algorithm based on density peak clustering and weighted similarity (DPWSAP) was proposed. The solutions were following: 1) densi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE access Jg. 7; S. 175106 - 175115
Hauptverfasser: Wang, Limin, Hao, Zhiyuan, Sun, Wenjing
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Piscataway IEEE 2019
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
Adaptive algorithms
Affinity
Affinity propagation
Algorithms
Clustering
Clustering algorithms
Convergence
Damping
Data points
Density
density peak theory
Economics
Euclidean distance
Finance
Human factors
Propagation
self-adaptive strategy
Similarity
weighted similarity
ISSN:2169-3536, 2169-3536
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To solve both the similarity calculation method and parameter limits problems of the affinity propagation algorithm (AP), the self-adaptive affinity propagation clustering algorithm based on density peak clustering and weighted similarity (DPWSAP) was proposed. The solutions were following: 1) density peak algorithm (DP) was introduced to create the local density attribute for AP algorithm; 2) weighted similarity was applied to heighten the similarity extent of data points; 3) growth curve function model was employed with setting a self-adaptive strategy for damping factor (<inline-formula> <tex-math notation="LaTeX">\lambda </tex-math></inline-formula>) to enhance the convergence performance of AP at different stages. To verify the performance of DPWSAP we tested six UCI data sets with different density, different dimensions, and data volume. Experimental results indicated that DPWSAP had better clustering accuracy and convergence performance than original AP algorithm and several other clustering algorithms. In addition, the self-adaptive strategy improved the overall performance for the algorithm, and reduced the possibility of human factors affecting the algorithm effect. The analysis results demonstrated that the DPWSAP had a good research value. Thus, the proposed algorithm had a better research prospect in theory and application fields.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2019.2956963