You’ve read a number of articles in this space about different types of machine learning, with a high-level view of how they work and the types of technologies that machine learning will enable in the future. “But,” you’re thinking, “What are some of the ways machine learning is being applied right now, to do useful work outside the laboratory?”
Funny you should ask. Even though real, useful machine learning is still in its infancy, there are some real-world problems that are being addressed by the technology. One of the early adopters is the financial industry. No, not all bank tellers, loan officers, financial planners, stock brokers, and CPAs will be replaced by robots anytime soon—but machine learning will help all of those professionals do their jobs better.
Financial Applications of Machine Learning
Here are some current or near-future applications of machine learning in the financial industry:
- Handwriting recognition. In the old days, if you made a deposit in an ATM, you put your cash and checks in an envelope, sealed it, and stuck it in a slot in the machine, where it fell into a basket. Eventually a human would take all the envelopes and process them, and when that was done your account would be credited. Many ATMs can now do the processing and crediting automatically, using image recognition powered by machine learning to identify cash and determine the amount on a check—even if it’s handwritten. For checks, you can skip the trip to the ATM altogether by snapping a picture with your phone. These systems are surprisingly good at accurately reading even the most wobbly handwriting. (So far, no luck making cash deposits by taking pictures of your tens and twenties.)
- Risk assessment. The brief heyday of “liar loans” is thankfully long gone. Loan officers have more rules to operate under, but also more tools at their disposal than just your credit score and old tax returns. Machine learning systems can examine your credit history and make even better, more accurate predictions of your creditworthiness than has ever been possible before. This goes not only for consumer loans but also corporate loans, which tend to have more complex risk factors.
- Fraud and intrusion detection. Financial systems are already using machine learning to identify unusual transactions in bank accounts, credit cards, and other financial activities. If you’ve ever been texted by your bank within seconds of having your ATM card denied because you were using it in a state you’ve never visited before, you can thank machine learning. Machine learning is also being applied to identify transactional patterns that are consistent with hacking. Soon, it will be unusual when a data breach is not found and fixed within seconds or minutes—as opposed to the weeks or months that is common now.
- Market prediction. Legend has it that J. P. Morgan (although some versions say it was John D. Rockefeller) was once asked his prediction on the stock market (or specific stocks—again, the legends vary). The reply: “It will fluctuate.” If Morgan (or whoever) had machine learning at his disposal, he might be able to supply a less flippant answer. Machine learning will be applied to identify patterns and trends in various financial markets and make recommendations as to whether to buy or sell (or even take those actions automatically on investors’ behalf).
When it comes to applying machine learning, the financial industry has a major advantage that many other industries don’t: It has tons and tons of current and historical data. This is a plus because all machine learning systems need to be trained, and the more data you can supply for training, the better the system will be in the long run. So it makes sense that the financial industry would be among the first to embrace the technology. Look for financial machine learning applications to expand in scope, complexity, and capability in the next few years.
