DDCM ： A Decentralized Density Clustering and Its Results Gathering Approach

Abstract

Distributed clustering is an important way to solve large-scale data mining problem. At present, there still exist some problems for distributed clustering, such as performance bottleneck of master node and network congestion caused by global broadcast. In the paper, we propose a decentralized clustering method based on density clustering and content-addressable network technique. It could generate a cluster just depending on the surrounding several nodes, which realizes load balancing and has excellent scalability. An optimization way of gathering clustering results in different application scenarios is addressed as well. Our extensive experiments show that the proposed approach performs 3 times better than benchmark algorithm in time efficiency and has a stable expansion ratio at about 0.6 for large-scale data.