Machine learning is a trending topic, but few really understand it. It’s becoming more important to the way we conduct business, and through its ability to process and interpret data on an unimaginable scale, it’s clear why. It starts with algorithms or a specific step-by-step process and series of calculations written out in code. It then uses a series of these algorithms to conduct specific acts or commands without the need for a programmer to tell it to. Its end goal is to solve a problem, define a pattern, or predict future outcomes. But where did machine learning come from?
The History of Machine Learning
We like to think machine learning was a result of science fiction movies and novels. Back in the day robots and talking computers were a figment of our imagination— not so much anymore. With the first computers came the possibility for machine learning. In 1959, a man named Arthur Samuel coined the term “machine learning.” Arthur was a talented computer programmer designing the first computer gaming interface with checkers. Ever wondered how you could play against a computer? You can thank this guy for it. Initially, machine learning was just a term for computational learning and pattern recognition, which still holds true today. However, with access to a greater volume and variety of data, and increased data speeds and storage, machine learning is finally living up to its potential.
Who or What does it?
Let’s set one thing straight; machine learning isn’t done by a machine, at least not in the traditional sense. Sure it needs a computer to do it, but it’s just a series of algorithms hosted inside a computer. Think of it like an evolving technology, where the most evolved machine learning processes become artificial intelligence. Computer engineers, software developers, and data scientists are all responsible for creating the basis for machine learning. While the possibilities can seem limitless, their job is to pair the right algorithms with the right data based on the tasks at hand– easier said than done.
How It Works
Machine learning, at a high level, collects data, processes it, illustrates it through a model, and then monitors and adjusts itself accordingly. Before beginning, machine learning first requires an objective, data, type of learning, and algorithm model. Objectives most likely come from management teams or decision makers. The data required will be set by the data scientist, researcher, marketer, or engineer. And while it’s a bit complicated to get into the algorithms and models, we can explain the types of machine learning with examples:
- Supervised learning identifies patterns in data based on pre-set factors. The pre-set factors include an input and hypothesized output. The input could be a woman who is 28 years old who purchased wine last week, and the output would be she will purchase wine again. The algorithm for this type of learning then collects data points based on the input and output and can then predict future patterns or trends. The output can look similar to a regression analysis.
- Unsupervised learning identifies patterns in data without any parameters put in place. For example, an algorithm is told to collect a specific set of data (i.e., demographics and purchase habits), then told to build relationships or patterns based off of that. A potential relationship could be that a 28-year-old woman purchases wine 2-3 times a month.
- Semi-supervised learning then identifies patterns in data when there are some pre-set factors and some unknown. Perhaps it’s known that 28-year-old women purchase wine, but the researcher wants to know when and how frequently.
- Reinforcement learning identifies cause-effect relationships based on experience with the data. Sticking with our same example, if a 28-year-old woman previously purchased a wine opener, then the data could reveal she will likely purchase wine.
In its simplest form, a machine learning process includes data collection, interpretation, and action or learning:
While there’s certainly much more that goes into implementing machine learning, this provides an introduction into machine learning. Regardless of the type of learning, machine learning is always an iterative process: as it’s exposed to new data it can adjust how it thinks. That’s why it’s been critical for data mining, statistical pattern recognition, and predictive analytics, specifically in marketing and market research— but more to come there!
Continue reading about machine learning and how it applies to market research by checking out the next post in our machine learning blog series. You’ll learn how machine learning applies to market research, the areas it impacts most and how, and the considerations to keep in mind when using machine learning in market research.