ALEXANDRIA, Va. —Whether at home or abroad, the main goal of any commander is to keep Soldiers safe. With that in mind, researchers at the U.S. Army Engineer Research and Development Center (ERDC) have been using large amounts of historical data, social media activity and news articles to identify the best indicators of when and where a terrorist attack may take place.
This “Big Data” approach uses massive amounts of available information, along with high-end parallel computing, to find important clues that aid in the fight against rogue activities.
“The threats analyzed can often come in the form of terrorist attacks or insider shootings and can take place both domestically and overseas,” said Ray Dos Santos, a mathematician and computer scientist with the ERDC’s Geospatial Research Laboratory. “You have to crunch millions of pieces of information trying to find a few interesting patterns. The idea is to calculate the probability of an attack in a certain area.”
The mostly automated system evaluates conditions that best indicate when and where such attacks may take place. One program attempts to identify rogue networks of terrorist activities eliciting malicious developments. Another line of research combines demographic data, geospatial events and street imagery to measure the likelihood of attacks in specific urban areas.
“We could be looking at a very busy street or a remote location,” said Dos Santos. “It could be an empty warehouse, a small desert area or something the size of a large army base.”
“Behind the scenes, the system looks not only into the landscape — the building layouts and population density — but also into the data,” he said. “For example, historical events that may have happened in that area, the area’s demographics, the population’s educational levels, income levels and housing. We make use of a lot of different information to calculate the score.”
This information is then computed and complied into a bigger picture. “It’s based on what we call graph neural networks,” said Dos Santos. “It creates little graphs of connections — it could be a place connected to a piece of data, connected to a piece of demographics in the area — and then it creates models of things that could happen.”
Getting an accurate picture depends on the amount of data on hand, and the biggest limitation of these algorithms lies in the lack of information or data.
“Depending on the location, if there is a lot of data and good pictures, the system works a little bit better than if it is a very remote area with a small amount of information,” he said.
Dos Santos says the team is running experiments with a prototype to increase accuracy.
“We want to make an impact at the lowest level possible,” he said. “We’re building these algorithms so that they can be implemented into existing systems — for example, mission planning. It would be nice if — before the Soldiers go out into the field ― we can help them better understand the area and know the full risks.”
“Threats can sneak up on you without any warning or without any suspicion whatsoever,” he said. “Our ultimate goal for this system is for our troops to be able to deploy into an area that they know to be safe.”