Improving results of mixture model based graph clustering methods using evolutionary algorithms

In the recent years there has been an interest within the physics community in the properties of networks of many types. Graph clustering is the process of identifying the network structure in terms of grouping the vertices of a graph into clusters taking into consideration the edge structure of the graph that in such a way there should be many edges within each cluster and relatively few between the clusters. Based on high computational cost, the classical algorithms will slow much since data size in real application increases rapidly. In such a situation, model based graph clustering algorithms are an efficient alternative to classical ones. The performance of the model based graph clustering algorithms depends on the correct initial parameter setting. We areproposedan evolutionary algorithm to find proper values for the model based graph clustering algorithms. The proposed method is tested on both simulated and real data sets and gave improving results in comparison with random parameter setting.