Swarming Geographic Event Profiling, Link Analysis, and Prediction

Geographically embedded processes with hidden origins are often observable in events they generate. It is common practice in criminological forensics to reverse simple equation-based models of trajectories that link the origin with its events to derive a probability estimate of a common origin location. This approach requires that linked events are manually extracted from larger event data sets. Also, as this approach is equation-based, it is generally not possible to take into account any specific geographic characteristics that may affect the trajectory. In this paper, we present a swarming model of simple geographic agents who reason "backward" from large sets of events to origin location probability distributions, use the overlap of these distributions to identify clusters of events that may share a common origin, and then reason "forward" from the clusters' origin distributions to predict the risk of future events. We apply this model to the domain of Improvised Explosive Devices (IEDs).