Cutting Through the Myths of Machine Learning
Recently, our Chief Science Officer, Dr. Pedro Bizarro, contributed to The Paypers’ “Web Fraud Prevention, Online Authentication & Digital Identity Market Guide” for 2015/2016 with an essay on machine learning. We find that with the fintech revolution underway, we sought to help bust several myths surrounding how automation can occur in analytics. While there are great opportunities for companies to get access to this new technology – especially to improve fraud prevention efforts in e-commerce – some may feel hesitant around the idea. Here are some common legends that just aren’t true:
Myth 1: Machine learning is only for big companies
Businesses may find the idea of new technology to be expensive simply because it’s only been recently out on the market. However, so much of machine learning is now possible because computing is now much more affordable. In the scope of 20 years, the cost of 1 million transistors on a single chip declined from $222 to just six cents. Storage, communications bandwidth and data sources also went down in price while expanding in capacity at the same time.
More importantly, machine learning doesn’t need to be a new program employees must spend weeks learning. We can now integrate these platforms as APIs and plug-ins. Simultaneously, we can use open-source code to customize a solution that’s affordable and specific to a company’s interest, including fraud prevention.
Myth 2: Machine learning takes away my ability to control my business
While the raw data that machine learning computes can be confusing and scary as it makes more decisions, there are now ways to better control and understand that information. The creation of semantic layers such as whitebox scoring helps create information that actually makes sense to the audience in question. In addition, the automation process frees up the fraud team’s time, giving a company more control over its resources.
Myth 3: I want the Uber-model that is best for all
The question a business should ask itself when they think of this is, “Do we have a single employee determining everything about a particular matter?” The answer’s likely no. Businesses perform best when there are team of different people working together. Similarly, ensemble methods of modeling ensures businesses get a coherent picture of what’s going on with their payments systems, eliminating biases and creating a consensus view of the situation.
Myth 4: Machine learning is all about the model
While the debate between better models vs. more data will likely last for many years, it’s important to remember a good model isn’t the only thing that makes a fraud prevention system effective. With constant attempts to find weak spots in the system a given, businesses should also bring in more data sources for the models to view. In this way, a robust defense against fraudsters can and will happen.
For more information on machine learning, check out “Myths about Machine Learning” in The Paypers’ “Web Fraud Prevention, Online Authentication & Digital Identity Market Guide” today.
Latest posts by amit sonawane (see all)
- Friendly Fire: Dealing with Deliberate Chargebacks from Consumers - May 19, 2016
- The future of payments and what it means for fraud - May 3, 2016
- The Internet of Things presents new vectors for payments – and fraud - March 24, 2016