Predicting Terrorist Actions Using Sequence Learning And Past Events

This paper describes the application of sequence learning to the domain of terrorist group actions. The goal is to make accurate predictions of future events based on learning from past history. The past history of the group is represented as a sequence of events. Well-established sequence learning approaches are used to generate temporal rules from the event sequence. In order to represent all the possible events involving a terrorist group activities, an event taxonomy has been created that organizes the events into a hierarchical structure. The event taxonomy is applied when events are extracted, and the hierarchical form of the taxonomy is especially useful when only scant information is available about an event. The taxonomy can also be used to generate temporal rules at various levels of abstraction. The generated temporal rules are used to generate predictions that can be compared to actual events for evaluation. The approach was tested on events collected for a four-year period from relevant newspaper articles and other open-source literature. Temporal rules were generated based on the first half of the data, and predictions were generated for the second half of the data. Evaluation yielded a high hit rate and a moderate false-alarm rate.