Current Issue
This Month's Print Issue

Follow Fast Company

We’ll come to you.

2 minute read

A Computer Program That Predicts Terrorist Attacks

A data-mining project that adapts analytics used by Amazon and big box retailers to predict customer behavior is now being used to stop terrorism before it starts.

South Asian terrorist organization Lashkar-e-Taiba (LeT) aren’t nice people. The group, which operates primarily in India and Pakistan, is best known for the devastating 2008 Mumbai bombings and for their long-term goal of a theocratic Muslim state covering all of modern Pakistan and India. LeT, which primarily operates in the disputed areas of Jammu and Kashmir, has also left an extensive paper trail that’s unusual for a terrorist organization. Now, researchers at the University of Maryland have created a massive computational analysis of LeT’s activities—that can even predict future strikes by the Lashkar. The analytic technologies that the University of Maryland team applied to terrorism are similar to data mining analytics commonly used by Amazon and big box retailers to predict customer activity.

The project, called the Computational Analysis of Terrorist Groups: Lashkar-e-Taiba, used algorithms to parse mined data on 770 variables from 20 years of LeT’s activities. These 770 variables were updated month by month for the computational analysis. Research was carried out by the university’s Laboratory for Computational Cultural Dynamics (LCCD) under the leadership of V.S. Subrahmanian. Information fed into the computational analysis system came from a variety of primary sources including open-source news articles from local magazines and newspapers, and scholarly publications on terrorism in South Asia.

Using the monthly data on the 770 variables, the researchers were able to establish an understanding of what factors determined how often LeT attacks occur, what different types of terror strikes are used in which geopolitical situations, what factors determine LeT’s use of proxy terrorist organizations for attacks, how LeT chooses attack locations, and many other criteria. Subrahmanian and researchers Aaron Mannes, Amy Silva, Jana Shakarian, and John Dickerson claim that the study determined that diplomatic pressure on Pakistan to rein in their use of terrorist proxy organizations and to disrupt LeT training camps is the most effective anti-terrorism strategy available for this situation.

The data mining project also yielded surprising results. "We were most surprised by the finding that arrests, raids, bans, asset freezes directed at [the organization] had very mixed success as a long-term strategy, as well as the finding that fostering dissent within LeT’s commanders is an essential part of destabilizing LeT," Subrahmanian told Co.Exist.

The proprietary software developed by the LCCD for this and other projects, called the Temporal-Probabilistic Rule System, received $600,000 in funding from the Defense Department. Similar computer modeling systems are currently on-record as being used by the governments of the United States, Great Britain, and India in anti-terrorism operations.

LCCD researchers have also applied machine learning techniques to the data sets in order to combat terrorism, based on the idea that algorithms could determine what conditions in the environment would be a good predictor of terrorist attacks occurring over the next few months. Using those algorithms, Subrahmanian found a number of "canaries in the coal mine" that could predict LeT attacks.

For instance, LCCD’s algorithms found that if between five and 24 LeT operatives had been arrested and LeT operatives were on trial in either India or Pakistan, there was an 88% chance of LeT attacking local security forces. Hundreds of other similar rules were generated, including predictors for terrorist attacks on civilians, professional security forces, transportation centers, security installations, and symbolic or tourist locations. Now you know what to watch for.