A study on cluster validity measures for clustering network data

Modularity is one of the evaluation measures for network data and used as the criterion of merging two clusters in Louvain method. To construct useful cluster validity measures for network data, the effectiveness of eight conventional cluster validity measures are compared with Modularity. Cluster partitions of six artificial network datasets are obtained by k-medoids and evaluated by cluster validity measures including Modularity. Numerical experiments show that the Dunn's index is effective in conventional cluster validity measures than other indices.