Patterning Out Terrorist Behavior
Industrial and systems engineering researchers at Binghamton University have devised a framework that could predict future terrorist behaviors by recognizing patterns and relations in past attacks. In a bid to learn how terrorist groups interact with one another, how they learn from what has happened and how the events relate to one another, Salih Tutun has amassed and used data from more than 150,000 terrorism and other criminal cases, with more than 70 features per event.
His framework included an evolutionary simulating annealing lasso logistic regression (ESALLOR) model and networked feature selection (NFS) method. The framework was able to predict various features of terrorist bombings in Iraq for the years 2014 to 2015. For example, geocoding specificity accuracy (whether the attack occurred in a city, village or town that had been attacked before) was 96 percent. Weapon type, the name of the attacking group, the vicinity, and whether the attack lasted less than 24 hours were all predicted with a 100 percent accuracy rate.
The framework has two main phases. The first builds networks for events, while the second uses a unified detection approach that combines pattern recognition approaches and the proposed network topology. A graph-based outbreak detection method defines hazardous places that have a potential for outbreaks of violence. The framework then calculates the relationship between selected features via a new similarity measure that can handle categorical and numerical features.
For example, Tutun’s advisor, Mohammad Khasawneh, chair of Binghamton’s Department of Systems Science and Industrial Engineering, was in Turkey recently when intelligence predicted an attack on French interests in that Middle Eastern country. They closed the French embassy in Ankara, Turkey’s capital, and the consulate-general in Istanbul on a Wednesday afternoon. A few hours later the news came about the terrorist attack in Nice, France, which killed 84 and injured hundreds.
Terrorist groups emulate the behaviors of others and learn from their mistakes and successes. Tutun’s framework helps model these changes and capture these learnings. Tutun still has work to do to expand the framework beyond Iraq as countries around the globe collaborate to deal with this problem in the interest of saving lives. But once the framework is verified and validated, leaders can take police reports, examine the statements, extract features and put them into the prediction engine, Khasawneh said.