Parallel Clustering Based on Partitions of Local Minimal-Spanning-Trees

Abstract

Many traditional clustering algorithms have the scalability problem while dealing with large data sets. One common strategy to handle the problem is to parallelize the algorithms and execute them along with the input data on high-performance computers. However, many graph-based clustering algorithms are hard to be parallelized since they need to calculate the similarity of all-pairs of all data nodes. In this paper, we propose a new parallel clustering algorithm, called the Para-CPLM (Parallel Clustering based on Partitions of Local Minimal-spanning-trees), which is based on three strategies - graph-based clustering, granular computing, and partition-and-merge. The Para-CPLM partitions the data domain into several regions for parallel execution, and then establishes a local minimal spanning tree in each region. After being established, the Para-CPLM combines those local minimal spanning trees and applies a method, namely the GBC method, to determine the best number of clusters. After the first phase of clustering, it repeatedly finds better pairs (edges) of the inter-clusters to reform the merged tree structure, such that the tree becomes closer to a global minimal spanning tree. Consequently, it uses the GBC method again to find the best number of clusters. From our experimental results, the Para-CPLM achieves significantly shorter execution time and better scalability while compared with the sequential GBC method. In addition, the clustering results are almost identical to those produced by the sequential GBC method.