Summary: | Affinity propagation (AP) clustering is a well-known effective clustering algorithm that outperforms other traditional clustering algorithms. However, the quality of clustering results depends considerably on related sensitive parameters (i.e., preferences and the damping factor). Thus, a feasible procedure based on golden section (GS) and the genetic algorithm (GA) is proposed. This procedure, called the “GS/GA-AP” algorithm, can perform proper global shared preference detection, including identifying a suitable number of clusters. A global shared preference is provided using the GS value between the minimum and maximum of similarities for AP as a default option, and the unsatisfactory clustering result becomes robust when the parameter with GA is selected. Finally, satisfactory experiments using one simulation data set and eight benchmark data sets are performed to verify the effectiveness of the proposed algorithm. The results indicate that GS/GA-AP clearly outperforms the original AP clustering algorithm.
|