Sequential pattern detection and time series models for predicting IED attacks

Abstract

Improvised explosive device (IED) attacks are a significant threat to coalition forces. Defeating IEDs as weapons of strategic influence has become a major objective of Combatant Commanders and their respective Joint Task Forces. This thesis attempts to identify new approaches that can help operational forces mitigate the risk of IED attacks by identifying common sequences of events that occur before an IED attack and forecasting the number of attacks in the immediate future. Using the CARMA association rules algorithm on historical data of religious, political, and IED attack events, a model is developed to explore commonly occurring sequences of events leading to an insurgency IED attack and to predict events that are likely to occur given the sequence observed to date. Time series models are also generated to identify trends and relationships that can be helpful in forecasting future monthly IED attacks based upon previous actual historical attacks. The identified sequences and forecasts could be used to help plan troop movements, rotations, force levels, as well as allocating limited resources to address imminent threats.