Supervised Learning: How AI Learns to Make Smart Decisions
Imagine you’re scrolling through your email and notice that most of the junk mail—those too-good-to-be-true lottery winnings and suspicious “urgent” messages—never even reach your inbox. You don’t think twice about it, but behind the scenes, something powerful is at work: supervised learning.
Your email provider has been trained on countless examples of real emails versus spam, allowing it to filter out the junk with impressive accuracy. Without this system, you’d be drowning in irrelevant messages, wasting time manually sorting through them.
The same technology shapes numerous aspects of your daily life—from your phone unlocking with facial recognition to Netflix knowing exactly what show to suggest next. But how does it actually work? Let’s break it down.
I wrote about machine learning here, but today we’re going to go into a different aspect of it. There are three main types of machine learning: supervised, unsupervised, and reinforcement. I’ll define each, but we’ll focus on supervised. Expect a post on unsupervised learning and one on reinforcement learning shortly.
The Basic Definitions of Supervised Learning
Supervised learning is where the model learns from labeled data. “Labeled data” requires an explanation too. Labeled data is a list of criteria that the model is going to use to make its determination.
In unsupervised learning, the model finds patterns in data without predefined labels. The algorithm acts independently to find patterns on its own. It’s a well-trained algorithm that can do that, but it doesn’t require additional interaction.
Reinforcement learning should be a fun topic to explore. In this method, the model learns by interacting with an environment and receiving either rewards or penalties. I’m going to be curious to find out what type of reward or penalty would be meaningful to a machine.
Why Do We Care?
Before supervised learning or any machine learning, was a thing, we got along without it. But there’s no doubt that there are things that are so much better now. Remember the early days of email? Or maybe today as well, if you aren’t using an email provider that offers a spam filter. Spam filtering is only one benefit of supervised learning. Some others are voice assistance like Siri, fraud detection that prompts the bank to shut down your debit card, medical diagnoses, personalized recommendations like Spotify’s suggestions, and self-driving cars (which I’m not ready to trust just yet).
Key Concepts of Supervised Learning
Supervised learning is one of the most widely used machine learning techniques because of its reliability and ease of getting started. It sort of works like human learning. A “teacher” provides examples – the labeled data – to help the model learn the correct “answers.” It’s a foundation for more complex AI models, like deep learning. (Yup, I just added that to my list of topics to cover later as well.) Some of the primary aspects of supervised learning are the features/labels connection, training vs. testing datasets, and generalization and overfitting.
Features (Inputs) and Labels (Outputs)
Features or inputs are measurable variables or characteristics that are used to make predictions. For example, if you wanted to predict house prices, features might include square footage, number of bedrooms, location, and the age of the house. Features can be numerical, like temperature and price, or they can be categorical, like color or brand.
Labels or outputs are the correct answer that we want the model to learn. If we want to train a model to be able to classify emails, the labels would be “spam” and “not spam”. In supervised learning, the model maps features to labels by learning patterns from the training data.
Training vs. Testing Datasets
A training dataset is the part of the data that is used to teach the model. The model fine-tunes its settings to minimize prediction errors. A testing dataset is a separate set of data that trainers use to evaluate how well the trained model can perform on examples it hasn’t seen before. The testing dataset helps to determine if the model is able to draw generalizations beyond the data used for training.
Generalization and Overfitting
Generalization refers to the model’s ability to perform well on material/data that it hasn’t been exposed to before. A well-generalized model learns the underlying patterns in the training data rather than memorizing specific examples.
Overfitting is when the model learns too much detail from the training data, making it too specialized. A good example of this is “teaching the test” to students. We see this often in Information Technology, when job candidates possess a certain certification but don’t really understand the job. The candidate memorized answers in order to pass the test but can’t do the job itself. Overfitting causes the model to do really well on the training data but not at all well on new data. There are specific steps that trainers can add to the training process, such as using more training data and cross-validation, that can reduce the risk of overfitting.
Types of Supervised Learning
Regression: Predicting Continuous Outcomes
Regression is when the output can take on any numerical value within a range. The goal of regression is to find relationships between the input variables and a numeric output. Weather forecasting is an example of this. We want to know the temperature for tomorrow, and that’s going to be the output. For inputs, we have the humidity percentage, the wind speed, the atmospheric pressure, the time of year, and yesterday’s temperature. The model learns how each of the inputs affects temperature, and makes a prediction.
Classification: Categorizing into Discrete Labels
When we use classification, the output is a category rather than a continuous number. The model assigns the inputs into predefined groups. Spam detection is an example of classification. Trainers teach the model that there are certain characteristics that spam emails have, but that a spam email must have a certain number of these characteristics. For example, just having the word “Urgent” in the subject line isn’t enough to call it spam, but “Urgent” in the subject line, as well as “claim your free”, plus a certain number of links in the body may be the criteria required to say, “Yup, that’s junk.”
Benefits and Challenges
You know by now that nothing is all good and very little is all bad. Most of our emerging technology is awesome and fun, but it’s all going to have limitations until some of the kinks get worked out. Let’s take a look at the benefits and the challenges.
Supervised learning models, trained on labeled data, often make highly accurate predictions for structured problems. This makes it work well for spam detection, medical diagnoses, and fraud detection, where there are plenty of labeled examples to learn from. Some of the simpler models provide clear insights into how they make their predictions. Supervised learning can be used in many real-world situations, like healthcare, finance, and technology. Businesses can use it to make data-driven decisions and improve automation. Once you get the model trained, it can quickly analyze any new data and make fast decisions.
It would be nice if it were all rosy, wouldn’t it? We know it’s not. Supervised learning requires a lot of labeled data – thousands, or even millions, of labeled examples. Labeling the data can take a lot of time and money, especially in fields where expert labeling is needed, like healthcare. If the data doesn’t have clear labels, supervised learning won’t work well, like a newly discovered disease where there isn’t any labeled data to train with. There’s also the risk of overfitting – if the model memorizes the training data, it won’t work with new data. Finally, training on really large datasets requires a lot of processing power, and that gets expensive
Your Turn
Now you know more about supervised learning than you probably expected. Is there a real-world application of supervised learning that you find particularly interesting? Drop a comment, and let’s talk about it!
Here are a few sites where you can dig deeper into supervised learning:
What is Supervised Machine Learning? | Elastic
Supervised learning – Wikipedia
Supervised Machine Learning | DataCamp
Machine Learning | Google for Developers
My photography shops are https://www.oakwoodfineartphotography.com/ and https://oakwoodfineart.etsy.com, my merch shops are https://www.zazzle.com/store/south_fried_shop and https://society6.com/southernfriedyanqui.
Check out my New and Featured page – the latest photos and merch I’ve added to my shops! https://oakwoodexperience.com/new-and-featured/
Curious about safeguarding your digital life without getting lost in the technical weeds? Check out ‘Your Data, Your Devices, and You’—a straightforward guide to understanding and protecting your online presence. Perfect for those who love tech but not the jargon. Available now on Amazon:
https://www.amazon.com/Your-Data-Devices-Easy-Follow-ebook/dp/B0D5287NR3
