Definition
Supervised learning is a machine learning method. You show the computer problems (data) paired with the right answers (labels) ahead of time, and teach it to find the answer on its own when it meets a new problem. It's called "supervised" because the teaching is like a person sitting beside you and showing you the answer sheet. By giving the computer many examples together with answers, it learns the hidden rules in the data by itself. It's the most typical way of teaching AI.
Studying with a Workbook That Has an Answer Key
Think about teaching a small child to tell cats from dogs. You show the child one photo at a time and say, "This is a cat," "This is a dog." After seeing hundreds of photos, the child sorts out in their own head what's special about cats, like pointed ears and round faces.
A computer learns in exactly the same way. You feed the AI lots of photos (input data), each bundled with a label, an answer tag that says "this photo is a cat." Then the computer analyzes the many pixel numbers that make up each photo, and starts to find the rules shared only by the cat photos.
After tens of thousands of rounds of practice, the AI can look at a brand-new cat photo it never studied and say, "There's a 99% chance this is a cat," getting the answer right on its own. It works like a student preparing for an exam who builds skill by working through a workbook with answers again and again.
To Be More Precise
To be more precise, supervised learning is the process where a computer searches for the most accurate math formula that gives the right answer (Y) when you put in an input (X). Widely used models like neural networks start with random numbers, so at first they spit out nonsense.
Every time the computer gives a wrong answer, it calculates the error (loss), meaning how far off it is from the real answer prepared in advance. Then it makes tiny adjustments to the many numbers (weights) inside its formula, so the error shrinks at least a little next time.
It repeats this process over and over to cut the error. But the goal isn't to push the error on the practice problems down to zero. That can lead to overfitting, which is like memorizing only the workbook. The real goal is to answer well on data it has never seen. In the end, the computer isn't memorizing problems and answers whole. It's finding the mathematical rules and patterns that run between problems and answers.
By Result, It Splits into Classification and Regression
The problems solved with supervised learning fall into two big kinds. The first is "classification." It's the job of picking one from a fixed set of choices, like deciding whether an incoming email is spam or normal, or which character a letter in a photo is.
The second is "regression." It's the job of guessing a continuous number, like predicting an apartment's sale price from the size of the room and its distance to the station, or guessing tomorrow's daytime high temperature from past weather records.
Supervised learning plays an important role in things we use every day, like face recognition on smartphones, auto-captions on video platforms, translators for many languages, and spam filters.
Common misconceptions
Supervised learning means a person sits next to the computer, watches in real time, and points out its mistakes.
A person isn't watching the whole time. The computer compares its predictions with "answer data that was labeled in advance" and cuts its own error by itself.
Where you meet it
A filter that automatically screens out spam, trained on hundreds of thousands of spam and normal emails together with answer labels
A quality inspection system on a factory line that learns from photos of parts and whether each is normal or defective, and automatically picks out the defective ones
In one sentence
It's a method where a computer learns from data that has problems and answers together, finding rules by itself and shrinking its error as it goes.