Two main Types of Machine Learning Algorithms
Machine Learning algorithms can be divided in two main categories - Supervised Learning and
Unsupervised Learning algorithms. The difference is simple but very important:
Supervised Learning algorithms: Let us suppose you are the owner of a car dealership centre.
Your business is growing, and you have to hire new staff and train them to help you. But there is a
problem – it just takes a glance to you at a certain used car to havea clear idea of how much it is
worth, but your new employees do not have a clue, since they do not have your experience in
estimating car prices. To help your employees (and perhaps get a few free days to go on vacation),
you decide to write a small application that can estimate the value of a car based on its
registration year, engine size, travelled kilometers and the average selling price of the same brand
of car. For 3 months, you register the price at which certain models of cars sell, writing down the
details of the transaction – year of the car, model, kilometers, power , etc., and last but not least
the final sale price. Using this training data, you can create a program that can estimate what could
be the right sales price of any other car of the same models (let’s say Mercedes and Audi). This is
called supervised learning. You know the price at which each car has been sold, so knowing the
previous answers to the problem, you are able to work backwards and understand the logic to
apply in order to solve new similar problems. To build your application, you have to feed the
Machine Learning algorithm with the data you collected regarding each car. The algorithm will try
to understand what kind of mathematical functions it must use to produce the solution for new
problems. Once you know what math function applies to a specific set of problems, you will be
able to produce a solution for any other problem of the same type!
Unsupervised Learning: Let us go back to our example with the car-dealership center owner.
How would you create the same application without knowing the sales price of each new car?
Even if you know only the kilometers and model of each car, you can yield interesting
results….this is unsupervised learning. It's like someone giving you a list of numbers on a piece of
paper and telling you, "I do not really know what these numbers mean, but maybe you can figure
out if there's a scheme or code or something at the base of them - Have fun! ".
So, what could you do with this data? To begin with, you might have an algorithm that
automatically identifies the different market segments in the data. You might find that Audis are
bought at higher prices below a certain mileage, but Mercedes’s can be priced like gold even
over a certain number of kilometers. Knowing these different types of customer choices, can help
you better manage your marketing work.
Another interesting thing you could do is to automatically identify any abnormal values that are
very different from all the others. Maybe Audis that are sold at abnormally high prices all are of a