Pattern classification in social network analysis: a case study

We present the methodology and results of a proof of concept study that characterized actors in a simulated dataset as terrorists or nonterrorists by applying statistical classifiers to their social network analysis (SNA) metric values. The simulated datasets modeled the social interactions that occur within Leninist cell organizations and those that occur in more typical social structures. Multivariate Bayesian classifiers operating on the actors' global betweenness centrality and local average path length achieved the best performance. These solved the three-class classification problem (cell leader, cell member, or non-terrorist) at 86% accuracy and the two-class classification problem (terrorist or non-terrorist) at 93% accuracy. An algorithm for defining local windows in multimodal social network graphs is also presented.