Using Machine Learning to Combat Digital Banking Fraud Part 1

Technology – The Ultimate Fraud Fighting Tool

Fighting fraud is something that virtually everyone can get behind. It’s a common good. This is especially true in the financial services industry where fiduciary responsibilities to customers and shareholders is paramount. Fortunately, technology, including machine learning, has evolved to make it easier for financial services companies of all sizes to accurately and cost-effectively identify suspicious activity indicative of fraud and halt it in time to avoid loss.

Machine learning is the data science that enables computers to learn from experience, as humans do. It deploys complex algorithms that automatically evaluate and analyze large data sets to find patterns. These algorithms enable computers to respond to different situations for which they have not been explicitly programmed.

Machine learning “teaches” computers to identify and distinguish the importance of patterns. It is found to be especially valuable when there is a lot of data, when time is an important factor, and when the wrong decision has a significant impact. For example, when making a decision about whether a particular financial transaction is legitimate. Machine learning allows these data-rich, time-dependent decisions to be acted upon more quickly and accurately.

In fraud detection, this means using algorithms to evaluate data and make a determination or prediction about the likelihood of fraudulent activity. Computers are more efficient than humans at processing large datasets. They are able to detect and recognize more subtle patterns. These algorithms are also adaptive and continue to improve over time with additional exposure to more data. This characteristic is often referred to as “self-learning”.

Fighting Digital Banking Fraud

One of the many ways that machine learning helps financial institutions combat fraud is by monitoring digital banking activity. By proactively tracking digital banking activity, new transactions can be compared to the expected behavior for a particular customer, based on a profile compiled from their historical behavior. The algorithms can identify a pattern activity. They can also calculate a fraud risk score based on how deviant the new activity is.

Some of the information that can be monitored to help identify suspicious activity includes:

  • User activity
  • Device fingerprint
  • Temporal data
  • Frequency
  • Velocity
  • Geolocation information.

Let’s look at a simplified example.

Sam Smith, a retail banking customer, typically logs in to his online banking platform between 8 am and 4 pm, Monday through Friday, from a laptop at his home in California.

Sam makes an average of eight bill payment transactions per month: 4 in the first week of the month, 4 in the last week of the month, with the average transaction amount of $1,800.

This is part of the data that makes up his “normal” behavior.

Then, Sam initiates six payments of over $2,500 each on the 12th of the month, from a mobile phone located in Toronto, Canada at 11 pm.

It is easy to see that this is not normal behavior for this customer. But this is just one customer among tens or hundreds of thousands initiating online banking transactions each day. Machine learning is what enables banks to manage the volume and complexity of these transactions and pinpoint what poses a true risk.

Guardian Analytics is thepioneer and leading provider of behavioral analytics and machine learning solutions for fraud detection and anti-money laundering software for financial institutions and enterprise organizations. Hundreds of financial institutions have standardized on Guardian Analytics’ innovative solutions to mitigate fraud risk and stop the sophisticated criminal attacks targeting retail, commercial, and enterprise banking clients. To learn more, click below to contact us and request a demo.