An Ensemble Approach for Clustering Scale-Free Graphs

Several real-world networks of interest, such as social and biological networks, are modular in nature. Most of these networks also possess the scale-free property, which makes the task of detecting and isolating communities from these networks difficult. The application of traditional clustering algorithms on these networks has not yielded a great deal of success. In this paper, we apply an ensemble clustering approach to address this problem. Ensemble clustering has been suggested as a good approach to improve the performance of clustering algorithms. To perform initial clustering, we employ specific topology-based distance metrics that are conducive for partitioning these networks. We use an ensemble of different base clustering algorithms to obtain a set of clustering arrangements. We apply consensus clustering algorithms on this set to obtain the final set of clusters. We demonstrate the effectiveness of the ensemble clustering technique on a real scale-free interaction network the Protein-Protein Interactions network of budding yeast. Our experimental results show that our approach can provide improvement over the base clustering algorithms. We also provide an empirical evaluation of different consensus methods proposed in the literature.