Applying Matrix Decompositions to Counterterrorism

Governments collect data in which they hope to find patterns of terrorist activity. It is hard to know what such patterns look like and, in any case, terrorists are actively trying to avoid leaving any distinctive traces. However, if they work as a group, it is impossible to avoid some correlation among their attributes and actions. We show that such correlation can be detected, partly because it is likely to be qualitatively different from the correlations among groups with more innocent purpose. We show that matrix decompositions, in particular singular value decomposition and semidiscrete decomposition, have several useful properties for this problem. In many cases it is possible to identify a terrorist group with few false positives.