Performance comparison of some centrality measures used in detecting anomalies in directed social networks

Abstract

Recently, researchers have shown an increased interest in combining Statistical Process Control/Monitoring and Social Network Analysis. One approach to detect anomalies in social networks is to monitor some summary statistics of the network structure using control charts. We aim in this study to conduct a performance comparison among some network centrality measures (betweenness, closeness, and degree) in terms of their ability in detecting anomalies using control charts. Although directed networks include more information on the nodes communication than the undirected ones, they are rarely considered and evaluated in literature. Hence, the performance comparison is conducted assuming weighted and unweighted directed networks. Two network-level measures are used; the average and standard deviation for each metric. Our simulation results revealed that the degree centrality measure in most cases is recommended to be used for both weighted and unweighted networks. Also, it can generally be concluded that the average of a centrality measure performs better than the standard deviation in detecting anomalies. However, if the structural change is major in weighted networks specifically, we recommend the standard deviation of the centrality measure to be used.