A Fusion Clustering Algorithm Based on Global Gravitation Search and Partitioning Around Medoid

Abstract

Partitioning around medoid (PAM) is a classical clustering algorithm widely used in practice, but it has some shortcomings such as sensitivity to the initial center and empirical determination of cluster number. To solve these problems, a fusion clustering algorithm based on gravitational search algorithm (GSA) and PAM is proposed. Firstly, regarding the clustering center set as the population particle, and the global search ability of GSA is used to optimize the initial cluster centers. Secondly, the population maturity factor is added to avoid the local optimum. Thirdly, the Davies-Bouldin Index (DBI) is added to evaluate clustering quality to obtain the optimal cluster number in the loop iteration. Finally, the simulation experiment is carried out on the four UCI datasets. The experimental results show that the fusion algorithm can effectively find the number of clusters in each dataset, and has higher accuracy and better ability to select appropriate clustering centers than K-means, PAM, GSA and the algorithm in literature [6].