Kxen discuss social network analysis - A new weapon in the fight against credit card fraud

When police raided 84 separate addresses across Europe, the US and Australia one day in June they weren’t chasing terrorists, drug barons or international smugglers. Rather the targets were members of an international credit card fraud ring.

It was the conclusion of a two-year investigation and led to the arrest of 178 individuals, members of a gang that had created more than 5,000 fake cards that - with the proceeds of other illegal activities - netted them more than $24 million.

But when taken against a global figure for card fraud of $6.89 billion in 2009 - and set to climb to $10 billion by 2015 according to the Nilson Report - the gang's efforts were little more than a drop in the ocean. Card fraud is now the number one fear of American citizens ahead of terrorism, viral infection and personal safety according to US publication the Unisys Security Index.

And as over-the-counter transactions involving credit cards get progressively more secure - thanks to technologies like chip and PIN - the focus of criminals is increasingly the online world where the volume of transactions is climbing all the time. From an estimated 420 million online purchases four years ago, industry analysts are predicting this will soon rise to 1.5 billion annually.

Cardholder not present (CNP) transactions now represent the fastest growing fraud risk in many parts of the world, with the majority of cases now directly attributable to organised crime. Such frauds are also a major source of income for terrorist groups around the world. 

Yet the current detection and prevention mechanisms used by the banks tend to be too slow to effectively fight this kind of fraud, often not highlighting problems until a card has been used several times. By then the money has gone and the fraudster has moved on to another card and another merchant.

Fraud Detection

For a long time credit card fraud detection has been rooted in data analysis techniques first developed in the 1980s, mostly with systems built around neural networks and decision trees.

While these can work adequately for cardholder present transactions where volumes are relatively low, they have not translated well into the online world where transaction volumes are much higher and speed of authorisation is everything. Nor do they offer any way of spotting links between apparently separate instances of fraud - one of the classic hallmarks of organised crime.

Compounding the problem is that banks do not want to block good transactions, the so called false positives. To do so would hurt business and potentially alienate good, credit-worthy customers. For this reason the tendency is for banks to let through a number of bad transactions, or false negatives, then rely on subsequent offline analysis to flag bad cards and block any further dealings. To stay one step ahead, all the fraudsters have to do is make sure they change their methods quickly and often.

Detailed, methodical analysis is more useful in criminal investigations. Police may be looking into many thousands of separate cases of card fraud, looking for common links and finding out whether the same individuals or groups are responsible. The group arrested in June being a case in point.

However, the sheer volumes of data involved, the relatively small number of fraudulent compared to genuine transactions, and the cost to the banks of blocking good transactions together conspire to make current detection techniques inefficient. But now a new analytical technique, social network analysis, has entered the war against credit card fraud. Augmenting the already established detection methods, it promises to help lower significantly the risk of card fraud.

Unravelling Fraud The Social Network Way

Rather than rely on the traditional way of mining historical data on transactions to detect, classify and predict fraud, social network analysis instead maps the complex networks of relationships that exist between frauds, cards, cardholders, transactions, merchants, industry sectors, phone numbers etc. If two or more of these network nodes share something in common - for example the same card used at two different merchants - then a connection exists. Add many thousands of nodes, plot all the connections between them, and soon patterns emerge that can help reveal fraud.

For example two cases of fraud may involve different cards and cardholders but the same telephone number or email address may have been used to make the fraudulent purchases. Further investigation may reveal other common factors with other cards (nodes) in the network and previously isolated groups - or islands - of fraud suddenly come together and start to make sense. It is being able to visualise these networks with hundreds or even thousands of nodes that is crucial to fraud detection.

In another case the same fraudulent card may have been used at two different merchants. When merchants and cards become network nodes it is possible to identify communities of cards and communities of merchants that are linked together, and which particular communities have a higher propensity for fraud.

With potentially many millions of nodes the task of mapping connections onto such social networks becomes massively more difficult, as does finding ways to navigate around them. The use of highly automated software tools can greatly ease the process and make results available in the minimum possible time.

Leading the field in social network analysis is the data mining automation vendor KXEN. Its social network analysis software module, KSN, is well proven with international telecommunications operators in combatting customer churn. Now the technology is behind an initiative to fight online credit card fraud with a major European financial consortium.

Early results show great promise and single figure percentage reductions in fraud are already being projected. With the scale of the crime as large as it is, even a small reduction can add up to tens or hundreds of millions of Euros saved.

Better still, the results of social network analysis can be used to augment the findings of the traditional analytical techniques used in detecting fraud. This can add significantly to the efficiency of an existing predictive model in determining which transactions are likely to be fraudulent.

As it is now acknowledged that the large majority of credit card fraud is committed by organised gangs - something which social network analysis is most effective at detecting - the case for social network analysis becomes all the more compelling.