DHS Funds Machine Learning Tool to Boost Other Countries' Airport Security

Passengers wait for pass the security control at Barcelona airport in Prat Llobregat, Spain, Friday, Aug. 11, 2017.

Passengers wait for pass the security control at Barcelona airport in Prat Llobregat, Spain, Friday, Aug. 11, 2017. Manu Fernandez/AP

The system would help countries determine if air travelers pose security risks without investing in expensive analytics software or personnel.

The Homeland Security Department is investing in machine learning technology that could help foreign countries increase airport security at zero cost.

The agency’s Science and Technology Directorate on Monday awarded $200,000 to Virginia-based DataRobot Inc. to begin testing an automated machine learning platform that would let airports assess the risks of individual travelers faster and more effectively. The contract comes as part of the Silicon Valley Innovation Program, an in-house startup accelerator for groups developing national security technologies.

If the prototype proves successful, Homeland Security plans to use the technology to enhance the Global Travel Assessment System run by Customs and Border Protection.

The open-source GTAS application is designed to help governments around the world that don’t have resources to assess flight risks boost their own airport security to U.S. standards. Using the program, data scientists can vet passengers and predict flight risks of individual travelers, which indirectly keeps unsafe people and cargo from entering the U.S.

“We believe there has to be a minimum bar, a baseline standard that people can use to assess the risks of travelers coming into the country,” said Anil John, S&T’s program manager for identity management research and development, in a conversation with Nextgov. With GTAS, countries can bootstrap advanced risk modeling while avoiding high upfront costs, he said.

However, the software today requires technical expertise, and the rapidly shifting nature of security means models are often out-of-date by the time they’re completed. But automated machine learning could speed up that process and simultaneously make it easier for non-data scientists to find the models that work best for them, John said.

Using DataRobot’s software, analysts can run models against each other to see which ones work best for certain predictions, according to John. As an add-on to the GTAS system, the platform could ultimately reduce the technical barriers to using the software and possibly compel more countries to adopt the technology, he said.

In the second phase of the four-part SVIP cycle, DataRobot will spend three to six months developing a working model of its prototype. Upon completion, the company would test its platform in a wide array of operational scenarios before going to market.

“With the number of international travelers to the United States increasing every year, we know we need better and faster tools to process incoming passengers,” said SVIP Managing Director Melissa Ho in a statement. “An enhanced Global Travel Assessment System will mean a better travel experience for all passengers and increased safety for Americans.”