Statistical Procedures for Forecasting Terrorist Organizations

• In Event: Perpetrators of Terrorism: Complexities, Comparisons, and Unknowns

Abstract

Attributing responsibility for an attack to a specific terrorist organization has generated growing concern among both policy makers and researchers. Approximately 47 percent of the attacks in GTD remains unknown, as little information on the identity of the terrorist organization was provided by the news articles. Furthermore, conflicting information and multiple unsubstantiated claims of responsibility make attribution challenging. Case-by-case systematic review of updated information of these complex cases is constraint by limited human resources. Confronting these limitations, our study attempts to improve understanding of this issue by developing a machine learning approach that proceed adaptively from the existing information captured in the GTD. Our study first compares the forecasting performance of 14 different classifiers on the top ten countries of terrorist attacks. Consistent with the machine learning literature, we find that the algorithms with the highest accuracy are decision trees, random forests and support vector machines. Next, we exploit the decision tree algorithm to forecast the unknown terrorist organizations for these countries. We conclude by discussing the implications of predicting unknown terrorist organizations for both research and public policy.

Authors

• Ugur Koc, University of Maryland

• Gary LaFree, University of Maryland

• Bo Jiang, University of Maryland