Demystifying machine learning

Machine learning is a phrase that gets tossed around quite a bit. If you ask three different people what it means, you may get three different answers. It is a bit different than your average buzzword, because it is actually a defined process.

I understand why it is on the minds of so many. Machine learning has lots of promise and applications for our industry. It is already changing the way we think about fraud detection.

The promise of machine learning is automation of complex manual processes. It is most useful for processes with a wide margin for human error. Manual review comes to mind here.

But machine learning is not something you can plug into your system and be all set. It helps to understand how it works and some ways it can be used to fight fraud. First, let’s cover what machine learning is.

Machine learning: definition and context

Machine learning refers to a type of artificial intelligence which provides computers with the ability to learn without explicit programming. It focuses on the development of computer programs that can change when exposed to new data.

By using algorithms that iteratively learn from data, machine learning allows computers to find hidden insights.

This may come as a surprise, but the concept of machine learning actually dates to the late 1950s. That is when a Stanford computer scientist theorised that instead of humans teaching computers, it could be possible for machines to learn for themselves.

More recently, the Internets ubiquity has given rise to rapid machine learning advances. The vast amount of digital information available for analysis is a perfect catalyst.

Machine learning and fraud prevention

For fraud prevention, machine learning has lots to offer. Primarily, it can help identify and more quickly react to risky patterns, behaviours and trends. The same pattern recognition is also possible for good transactions. In practice, this allows streamlined approvals of legitimate transactions with minimal friction.

Machine learning represents intelligent automation of up front fraud risk assessment. It can optimise the building and adjustment of rules around observed fraud tendencies. This approach enables growth, offers greater scale and resource optimisation. For merchants, this is especially important, as ecommerce demand only continues to rise.

Machine learning is not the final frontier for fraud prevention, though. The concept of deep learning takes machine learning to the next level. Though somewhat new, deep learning offers many potential benefits for fraud prevention teams.

Deep learning

Deep learning refers to a specific class of machine learning and artificial intelligence. It is based on so-called neural networks, which correspond to a class of machine learning algorithms.

First, some history. The development of neural networks dates to the 1950s. Inspiration came from the human brains use of pathways to interconnect information.

How deep learning works

Deep learning groups artificial neurons into layers. Information flows unidirectionally. In a layer, each neuron communicates with the rest until reaching the end of the network. The result an ability to power a computer system using a huge amount of data for complex decision making.

Deep learning for fraud prevention

Deep learning lends an extra layer of accuracy to existing machine learning models. In my experience at Emailage, deep learning holds lots of value for fraud prevention.

The chief benefit is the reduction of false positives. In an age where customer experience is a priority, this method has huge potential at scale.

Our core product is email risk assessment. We examine the behaviour and history associated with an email address to generate a predictive risk score. As our product continues to evolve, we have adopted technologies to power some of our latest features:

Element Historical Behavioral Monitoring (EHBM): Builds historical profiles for all transactional elements provided to identify normal customer behaviours.

Instant Abnormal Pattern Detection (IAPD): Real time out-of-pattern detection via comparison of current activity against historical profiles (EHBM) of customer transactions.

Customized Machine Learning: The ability to deploy unique machine learning modules at customer level, with rapid updates.

Deep Linkages: Enhanced graph database which correlates transaction elements for more robust risk assessment.

A few years ago, these technologies would be considered fraud prevention science fiction. It is very exciting to work in a time when they are coming online and providing real benefits.

Fraudsters do not quit. But these technologies, when properly used, give fraud pros a big advantage in the fight against them. Many disastrous trends will be stopped long before going mainstream. I look forward to working with companies in every industry to make sure this is the case.

About Amador Testa

Amador is an industry expert in online fraud, identity theft and cybercrime. Before Emailage, he was the head of fraud for card acquisitions at American Express and later led global fraud prevention divisions at Citigroup. Amador enjoys playing tennis, running marathons and traveling with his family.

About Emailage

As the global hub of email intelligence, the Emailage team has a singular goal: harnessing the power of the email address to help our customers achieve the delicate balance of reducing fraud levels while delivering optimal customer experience. Headquartered in Arizona, USA Emailage also has offices in Brazil and the United Kingdom.