Evaluation of Terrorist Scenarios Linguistically Using Belief and Plausibility

The risk from a terrorist scenario can be expressed as a function of three variables: the frequency of the scenario, the probability that the scenario is successful, and the consequences of the scenario. A scenario includes: the target and desired consequences, the resources (including information) used by the adversary, and the attack plan. The frequency of a particular scenario is determined by the adversary evaluating that scenario in the context of many other potential scenarios. The process by which the adversary selects a scenario (i.e., determines it is more “attractive” than others) involves consideration of the types and magnitudes of the potential consequences, the resources deemed necessary to defeat any security perceived to be in place, and the ability to gather the resources required without detection. The defender attempts to “think like the adversary” in evaluating potential scenarios, but - for the defender- there is considerable epistemic (state of knowledge) uncertainty as to the scenarios that may be selected by the adversary. The adversary has a choice of scenarios, so the evaluation of the likelihood of a particular scenario is more concerned with the reasoning process by which variables of concern are deemed “attractive”/“acceptable” by the adversary rather than a numerical evaluation of the variables. The attractiveness of a scenario can be modeled linguistically with each variable of concern segregated into fuzzy sets. The linguistic variables are combined using an approximate reasoning rule base on the fuzzy sets, where the rule base reflects the decision process for evaluating the scenario. To capture the epistemic uncertainty inherent in attempting to “think like the adversary”, the belief/plausibility measure of uncertainty can be used; evidence is assigned to families of fuzzy sets for each variable. To facilitate the evaluation of terrorist scenarios in this way, a linguistic evaluation methodology using approximate reasoning and belief/plausibility has been implemented in a Java-based software tool called LinguisticBelief. A graphical user interface allows creation of variables, fuzzy sets, and the reasoning rule base. Focal elements can be assigned for a particular scenario and the belief/plausibility interval for the fuzzy sets can be calculated for any variable in the model. Scenarios are ranked using the complementary cumulative plausibility distribution, and sub-ranked using the complementary cumulative belief distribution. This approach has been applied to cyber security, security of nuclear materials, and security of nuclear weapons sites.