The Department of Immigration and Citizenship will deploy a new risk tiering system at all international airports to improve its ability to identify “problem travellers” in real time.
The so-called border risk identification system was built and tested in-house over four months, using the open source R environment and about eight years of historical data.
A prototype was deployed in Sydney, Melbourne and Brisbane airports in late January last year, powered by two Dell desktops.
DIAC’s intent management and analytics director Klaus Felsche said the new system had halved the number of travellers undergoing additional checks at airport immigration points.
The department has $1 million in Australian Government funding to integrate it with border control systems in production at all international airports by March next year.
“Before we did this, we were pulling 2500 people every month in Sydney Airport alone off for additional checking, using what were business-type rules [that were] manual to a large degree,” he said.
“[Of those, we had] 55 cancellations. What that means is we had sufficient evidence to actually cancel a person’s visa – which is a lot harder than refusing one – and sending them back at the airline’s expense.
“Now we’re pulling up about 1200 to 1500 a month, so around half, and we’re getting over 60 effective refusals.”
Felsche said each of the refusals saved the Australian Government $60,000 on average.
He told the CeBIT Big Data conference last week that DIAC was sitting on a “platinum mine” of data, with records of more than 500 million border crossings and gigabytes of unstructured intelligence.
DIAC had two data warehouses – a legacy Oracle system and a newer DB2 system – that fed a third sandboxed data store.
Analysts used the third store, as well as records management (TRIM) and intelligence (IMtel) systems to build and test models without affecting the integrity of warehoused data.
Felsche noted that the department was unable to tap into much of other departments’ data stores and social media due to Australian privacy laws.
In January, two friends were detained by US authorities at the Los Angeles International Airport after one tweeted that he would “go and destroy America” – when he reportedly meant that he wanted to party.
“We have to deal with privacy issues and data management issues,” Felsche told iTnews of the challenges of using social media as an input into DIAC’s risk calculations.
“We have to build systems to store data, potentially integrate it with data that could be quite sensitive, particularly if it’s linked with immigration records.
“So the aim is yes, we would, but not yet. I’ve got much more core, core business to deliver against first which is easier and within my reach.”
Building without a budget
Felsche credited DIAC’s investment in people, not systems, for its ability to build and test the border risk identification system at relatively low cost.
DIAC analysts initially obtained input data from files that were updated hourly in a group drive. The system returned predictions to airport staff through Excel spreadsheets.
The analytics engine initially ran on three Mac desktops – with a total of 32 processor cores – purchased for a total of $9000 and networked through the Snowflake library in R.
After weeks of having the Sydney Airport “bombarded with 48 Excel spreadsheets every day”, Felsche purchased an off-the-shelf version of database application FileMaker Pro to improve user experience.
Analysts visited airport staff to fine-tune the dashboard. DIAC also consulted with the Commonwealth Bank, Australian Taxation Office, Centrelink and New Zealand Government on its models.
By the time Felsche put forward a formal proposal for funding to the department in May last year, the prototype had already won the support of staff on the ground.
“If I had gone to some of the vendors and said, ‘Give me a solution’, by the time we’ve finished the proposal paperwork and tendering process, we wouldn’t have had any money to buy anything,” he noted.
“If you’re smart about it, almost anybody can do what we’ve done; I wanted to invest in smart people rather than invest in platforms.”