Although it is only one element of Artificial Intelligence, the vast majority of commercial applications of AI are based upon some form of Machine Learning. But how does Machine Learning actually work?

Although some experts try to shroud it in mystery by using technical jargon, it’s actually fairly straightforward. Machine Learning basically automatically finds the “best” fit for a model by looking at “training” examples: you could do the same thing manually, but Machine Learning saves you the bother.

To take a closer look, let’s look at the simplest Machine Learning algorithm : Linear Regression. Anyone who has ever done a Statistics course will be familiar with Linear Regression. Let’s take an example of how it’s used.

Imagine you are trying to build a model to predict house prices in a particular area. To do this, you might first of all brainstorm on all the things that might affect the price: size, location, number of bedrooms, number of bathrooms, age etc etc.

You would then try to figure out how much each factor affects the price. To do this , you would look at all the data you have on houses in the area of interest and try to “fit” the best model possible. A “Linear” model would assume that the price is proportional to some linear combination of all the various factors that you have identified. The model that gives you the closest fit to all the data you have reviewed is considered the “best” fit.

Once you have created the model, you might test it on any new data to see if how accurately it predicts the house price.

So, how would Machine Learning handle this same problem? Well, in almost exactly the same way. You still need to give it the factors (or “features”) that you think are important for predicting house prices. Plus, you need to give it all the data you have on house prices (this is called the “training data”). If the Machine Learning algorithm is using Linear Regression, it assumes that the best model is a linear combination of all the factors you have identified and automatically “learns” the best fit to the training data. It then uses this learned model to predict prices on any new data you give it.

Of course, this is somewhat of a simplification and there are much more complicated algorithms than Linear Regression. But in essence, all Machine Learning provides is an automated way to find the “best” fit for a model from a large amount of training data.