Padlock / code / hacking

How machine learning is improving fraud detection

With fraud now costing businesses $3.5 trillion a year, many businesses are turning to AI-powered machine learning for help. Take the case of Nordic Bank, which was struggling to fight fraud using traditional methods. After adopting machine learning fraud detection, false positives fell 60 percent and true positives rose 50 percent, allowing Nordic Bank to spend more time solving actual fraud cases instead of chasing false leads.

Machine learning is helping companies like Nordic Bank make major strides in detecting and preventing fraud. Here’s a look at how machine learning is helping protect companies and consumers from fraud.

Making fraud detection more efficient

Machine learning makes fraud detection more efficient by using the latest artificial intelligence advances to uncover fraudulent patterns that wouldn’t be detectable using traditional methods, explains analytics provider SAS. Traditional AI fraud detection methods rely on fixed rules to define fraudulent activity so that it can be spotted when it occurs. This works well as far as it goes, but it’s only effective when thieves behave according to the rules. When criminals develop new methods of perpetuating fraud, traditional methods become ineffective.

That’s where machine learning comes in. Instead of relying on set rules that dictate predictable outcomes, machine learning applies probability theory to analyze patterns in data, matching a variety of mathematical models to the same data set in order to identify which pattern yields the best approximation to the data. By using probability to “learn” from data, machine learning can analyze new patterns that don’t fit set rules. This allows machine learning to identify new fraudulent tactics, make predictions about whether or not a particular behavior pattern represents a genuine fraudulent incident and recommend actions based on whether a given pattern has been identified as a real fraud incident.

Detecting online fraud

One way machine learning is being applied to fraud detection is making online transactions more secure. Given the current rate of identity theft, companies that process online payments must take measures to spot and block suspicious purchase attempts. Machine learning is well-suited to identifying suspicious online purchase behavior and intercepting it.

For instance, PayPal has over 170 million customers and processes $235 billion of transactions a year. By applying machine learning, PayPal can detect warning signs that a PayPal account is being used by someone other than the legitimate cardholder. For example, if the account of a U.S. customer is used by someone in China, PayPal needs to quickly assess whether this is a sign of identity theft or simply a U.S. citizen on vacation. By analyzing the customer’s purchase history and drawing information from fraud patterns that have been previously detected, PayPal’s machine learning algorithm can make an initial assessment of whether this is a legitimate transaction or whether it should be flagged as suspicious. If the transaction appears suspicious, additional security measures are implemented to prevent identity theft. Similarly, PayPal can tell the difference between whether a group of friends are buying concert tickets together or an identity thief is purchasing tickets using multiple stolen accounts.

Protecting devices

Another way machine learning is preventing fraud is by protecting user devices. When devices get infected by viruses or malware, the potential for fraud escalates. Stolen devices can also lead to fraud.

To stop fraud tactics that target devices, leading mobile device manufacturers have developed technology that harnesses machine learning to the task of on-device fraud detection. For instance, mobile processor manufacturer Qualcomm has developed an AI platform that has machine learning built directly into mobile devices in order to detect previously unknown “zero day” malware threats. The platform also uses machine learning to support biometric facial recognition in order to prevent thieves from using stolen devices to make fraudulent transactions.

Machine learning takes automated fraud detection to the next level by enabling detection of previously-unknown attack patterns, outsmarting thieves’ attempts to fool computers by changing their behavior. This allows online transaction providers to detect fraudulent behavior, as well as empowering mobile devices to identify suspicious activity. By protecting devices and online transactions, machine learning helps thwart the efforts of identify thieves and keep consumers and businesses safe from fraud.

Photograph by Typography Images

Share This