Machine Learning: Discover Your Problems Before They Happen

Dhanum Nursigadoo
5min read

There’s a sad truth we have to accept around AI and Machine Learning (ML). Right now neither are going to end with the sci-fi films that we all know and love. Most people hear AI or ML and immediately have that in mind, but we’re a long way off from robot companions taking over the world. The global proliferation of Human-Like Intelligence is unlikely to happen anytime soon.

So, what do people mean when they talk about AI and ML? AI is a branch of computer science that focuses on all simulated intelligence in computers; Machine Learning is a subset of that. At its simplest level it’s a labelling device. That is vital in speeding up processes you already have in place and freeing up time for employees. ML is brilliant at taking a task or series of tasks involving a huge dataset and completing them faster over time. It can eventually identify what a normal transaction looks like and use that information to detect any anomalies that exist in your data.

First things first, apart from the academics and sci-fi fans, AI/ML are currently used interchangeably. It’s an automated process of labelling things by computers to sort your data into meaningful information that you can use. It’s a hard thing to accept when so many people believe that AI means a computer assistant or a chatbot that can solve your problems with a simple question, sadly it’s not quite that convenient just yet.

What is Machine Learning?

But that doesn’t mean you should write off the benefits of AI and ML. Machine Learning is the application of AI, giving machines access to high volumes of data with a description of what we want and letting them figure it out for themselves. ML is a thing-labeller. That’s the simplest description of a complex topic. It takes your description of a thing with the label you want assigned to it, then it matches that description to the data you input. Hopefully, resulting in your data being sorted in a way that helps you.

Machine Learning is a technique used to train predictive models based on mathematical algorithms; it analyses the relationships between data fields to predict unknown values. Essentially, Machine Learning lets computers use data you already have to forecast future behaviours. And the best part is, they learn to do it themselves.

That’s where the real power of Machine Learning comes into play. It gets better and better at identifying what you want over time. It finetunes the description of what you want by understanding what you don’t want.

Who’s Actually Delivering Machine Learning?

AI and ML have been buzzwords for a long time across industries. But right now Microsoft is leading the industry by delivering meaningful ML solutions for enterprises through Azure. Part of that is because they’re specialists in the field, with having developed an entire studio dedicated to Machine Learning. The studio allows for easy predictive analysis based on your own data sources through a simple drag-and-drop interface.

Compared to more user-complex ML systems available, having an effective visual interface with Machine Learning algorithms that slot into your existing data centre is invaluable.

How ML improves data usage and use cases

So, how can ML work to help you? Anomaly detection is the clearest applied usage case right now. It’s already being used to detect credit card fraud and credit risk.

For banking, Anomaly Detection is one of the most salient use cases. It’s used to discover any outliers in your datasets that extend outside of normal usage. For credit cards that could be unusual spending and risky applications. Machine Learning lets you create a set of rules, a mapping function, around what you want to label. For example, rules to define what makes a ‘good’ application. Then the computer filters out everything that doesn’t fit those rules to develop a target function. One recent case of this was discussed on our Fintech Insider news podcast. Monzo discovered Ticketmaster customers were suffering from anomalous transactions following a data breach for the latter; within four and a half hours Monzo, who has used Machine Learning for over a year, was able to block future similar fraudulent charges.

Humans cannot create a target function. The mathematics on it are ineffable. Machine Learning is so useful because it creates the target function. It figures out what you want based on the broad rules around the mapping function.

Two Types of Learning

However, supervised Machine Learning bumps up against its limitations here as the mapping function needs recalibrating by humans to better inform the machine as to what the target should be. This is supervised learning for when you know what you want from your data, but the mathematics to find it yourself just aren’t possible.

On the other hand, there’s unsupervised learning. When you have a dataset with no firm decision on what output you want, or there’s fear of human bias, then unsupervised learning is a method of delivering useful patterns that you were unaware of. You can use Machine Learning to solve the underlying jobs to be done at the core of every banking enterprise. However, both unsupervised and supervised Machine Learning have a significant limitation in common. Both require massive datasets, literally millions of examples, before they can deliver real and effective outputs. They also both require skilled human interaction to orchestrate outcomes that are either fine-tuned to your needs or to rerun your data until it provides a useful output.

Forecasting is another tool available to you which is supercharged by Machine Learning. You can deal with customer-centric issues as they pop up to discover if something positive is going to happen. It provides researchers and compliance officers with real time intelligence to detect things before they happen. It has access to greater datasets and can process them faster than traditional models. Neural networks provided by Machine Learning beat traditional time series forecasting every time. However, these two systems can be made to work together to deliver more granular and accurate forecasting than ever before.

Why Azure does Machine Learning best

ML allows you to free up employee-time by automating data-grinding work. That means your highly-trained staff are now able to focus on what they’re best at. Taking data outputs and applying them to your business in a useful way. ML augments what you already have by providing you with a bevy of upgrades to enhance the human element of your business.

Azure is adept at seamlessly providing open and elastic Machine Learning solutions to make your data provide more insight than ever before. Azure ML brings out the power of the cloud for Banking and Capital Markets through a set of tools that are always accessible to you no matter where you are.

Right now AI/ML might only be as intelligent as a 4 year old. But before long it’ll be as smart as a PhD, and after that who knows where it will take us. Eventually your staff will focus on bigger issues than crunching the numbers to find actionable data.

Don’t forget to follow and on @Azure @IndustryXp and @msftfinserv on Twitter to stay up to date with all the latest news from the Azure team.

Keep an eye out for another blog post in association with Microsoft Azure in the future and an insights podcast on AI this Friday.