Terror prediction isn't quite like weather prediction. There are no warnings about widely scattered showers of shrapnel or heavy bombing, tapering off toward evening. Rather, counterterror data mining is being used to suggest high-level strategies for dealing with major terrorist organizations.
At the University of Maryland's Institute for Advanced Computer Studies (UMIACS), researchers have developed the SOMA (Stochastic Opponent Modeling Agents) Terror Organization Portal, or STOP, to help predict how terror groups will act and to share data through social networking.
"What it does is it takes a situation that someone might hypothesize or might be true, and tries to predict how a group might act based on what has actually happened," said V.S. Subrahmanian, computer science professor and UMIACS director. "We actually achieved accuracy on the order of about 90-plus percent ... but the predictions are somewhat coarse grained."
STOP won't tell any of the four defense agencies using the system that an attack will take place tomorrow, said Subrahmanian. But it will show that a specific group is likely to, say, increase its reliance on suicide attacks under certain conditions.
"It's more useful as a statistical tool to tell what a group will do," said Subrahmanian.
Created in collaboration with computer scientists and political scientists, STOP uses decades of data on the behavior of 30 major terrorist organizations, including Hezbollah, Hamas, and Hezb-I-Islami, to predict future behavior from past actions and events.
"As an example, you would think that when organizations are democratic or engaged in the political process, that they'd be more peaceful," said Subrahmanian. "However, that's not true with Hezbollah."
Subrahmanian said that STOP, based on 20 years of data, showed that when Hezbollah was engaged in electoral politics, the likelihood that it would attack Israeli civilians during an unspecified period ranged from 69% to 87%. That percentage declined to a range of 41% to 59% when the organization was not actively engaged in politics.
"This raises a lot of strategic questions about them," said Subrahmanian. "Does participation in elections actually increase the violence they're involved in?"
Such information has obvious value in both the political and military arenas.
"We think of this as an analytical tool that will help decision makers and policy makers," said Subrahmanian. "So a large portion of this involves a social networking component as well."
Subrahmanian, however, acknowledged that there's an inherent challenge in sharing information when some of that information may be sensitive or classified. For that reason, he said, STOP only includes open source data at the moment.
Awareness of STOP as a predictive tool naturally raises questions about whether the system could be tricked. Subrahmanian considers the possibility unlikely. "I think certainly there's always a tendency to game the system," he said. "You look to see what the other guy is doing. But the type of data mining we're doing is not something that's going to be immediately apparent to everyone."