Dynamic Bayesian Network for Predicting the Likelihood of a Terrorist Attack at Critical Transportation Infrastructure Facilities

Abstract

This research is motivated by the increased awareness for terrorism related research in the post-9∕11 era. Transportation infrastructures such as airports, metro and subway systems, bridges, and tunnels are vital to the U.S. economy. Therefore, in the wake of recent terrorist incidents, innovative and robust methods need to be exploited to predict the likelihood of terrorist strikes at critical transportation infrastructure facilities. Since the greater need for predicting future terrorist activities has only been recognized in recent years, the research in this field is in the nascent stages. There are two key aspects to predicting future terrorist activities: (1) developing a reliable and robust prediction model; and (2) analyzing the precision and reliability of available intelligence and other relevant information needed to develop a prediction model. This paper focuses on the first aspect and develops a terrorist attack prediction model (TAPM) using dynamic Bayesian networks (DBNs), which can be used to predict the likelihood of future terrorist activities at critical transportation infrastructure facilities. Theoretical development of the TAPM is presented and the model is employed in two examples to predict a terrorist strike with the possibility of an airplane hijack at a typical U.S. airport. The results suggest that the proposed DBN approach, although more data intensive, may provide a more reliable and better prediction. Since the nature of terrorist strikes is different than natural strikes, a game theoretic approach may be suggested in future works. Moreover, considering uncertainty in data as well as possibility theory in lieu of probability theory are also suggested for future research. This research is the significant first step in terrorist activity prediction at critical infrastructure facilities. Future enhancements to the model for more reliable predictions using real-life attributes are discussed.