Fusing Transactional Data to Detect Threat Patterns

The collaboration required for terrorist activities leaves a trace: a small, threatening pattern of transactions concealed in a background of transactional "noise." An approach to fusing transactional noise in order to detect the underlying threat patterns is outlined, applying ideas from detection theory to the theory of networks. A particular noise model is introduced: the random collaboration model, which models the transaction graph as a projection of a hidden bipartite entity-activity graph. This model is studied in some depth, as it sufficiently realistic to model real-world networks with good fidelity, yet sufficiently simple to allow to exact theoretical results and efficiently computable approximations. These approximations allow one to assess the likelihood that observed transactional patterns arise from some threatening activity, or merely from noise