Machine learning. Another buzz-word. If you’re not working directly with it, chances are you feel a bit insecure when it comes to the exact definition. Yet, as always, everyone else seems to have catched up. You also have an unpleasant feeling that it’s much closer to your everyday life today than just a couple of years ago. How embarrassing. Don’t worry though. This is your quick go-to guide for an introduction to the field. Soon you’ll be able to have intelligent conversations about the field, independently if you’re working as a project manager, quality assurant, developer, executive or just have an interest in tech. Hopefully, you’ll also understand the big value.
In order to get a feeling for what we’re talking about here, we could simply look in to the most essential process that occurs and creates value within the field. Ever wished you could look into the future? Humans have throughout the history found various ways to do this, with a variance in result and scientific proof connected to the approaches tested. With a quick look at the timeline where humans have existed, it isn’t until modern times we can do more general predictions with a higher certainty. For instance, one of the most estimated mainstream predictions done by science today is probably about the weather. This is the result of many years of research within the field of meteorology, oceanography etc., which gives a pretty good knowledge base for how the weather behaves. These ‘rules’ of how the weather behaves could then be used in order to predict the future. As you can tell, this process is solely based on humans own perception. Thus, it is also limited to the capacity of the human brain. What if there’s other ways to achieve knowledge, maybe in a way which is pinpointed towards an organizations own specific interests?
There you have it, the concept of machine learning, what it’s good for. But how’s it done though?
Nowadays we collect data about everything. This is of enormous value for organizations if analyzed in a correct way. Machines could be used in order to identify patterns in the data. With this approach, we could analyze data in much higher volumes than before. The patterns could then be used in order to predict the future. This could be applied to any area of business as long as we collect data regarding it. Identifying patterns is pretty easy. However, identifying the correct patterns is more of a challenge. Correct patterns are of course the only thing that’s applicable as a base for a real world estimation, independently if we’re trying to offer the right products to a customer, identify security risks or make money on the stock market.
Ok, so you know what you’re trying to predict. You also have enough data, providing a history for similar situations and their outcomes. How do you do it then?
In this case we’re just going to assume that the data we have contains relevant patterns. This data could then be fed to a machine learning algorithm. What this algorithm actually does is finding the patterns that’s hidden in the data, something that would take up a lot of time, if not impossible for a human mind to perform. The presence of high amounts of data, as well as computing power with a high capacity, is some of the factors that makes machine learning so relevant today. The recognized patterns then result into a so called model. The word is used in the same meaning as when creating a smaller instance of something to understand the reality better. Real world data about a current situation could be given to the model in order to get a prediction about the outcome.
Let’s look into an example to be more concrete here. Let’s say we have a phone operator. We want to predict when a customer is going to switch to a competitor. Using our collected data regarding customers who have cancelled their subscription, it’s possible to find patterns by providing that data to the machine learning algorithm. The result is a model explaining situations where customers might switch to a competitor. In order to try to keep these customers, the company wants to give them a better offer. Our application that handles all the marketing towards current customers can then ask the model, based on current user data, which customers that are likely to switch to a competitor and thus qualifies for a better deal.
This is what happens on a very high level. Yeah, seem simple huh? All the parts in the diagram is of course simplified. The processes of collecting the right data, create algorithms, build relevant models and so forth is quite complex. However, the results achieved when done right is very exciting. We’re still probably just looking at a glimpse of what the future has to offer.