In plain English
Supervised learning is a machine learning approach where the model trains on labelled data, that is, input-output pairs, so it learns to predict the correct output for new inputs. The labels act as the answer key the model learns from. It is the most common form of machine learning and underpins most classification and prediction tasks.
Why it matters
Supervised learning is powerful but hungry for labelled data, which is why labelling quality and cost are often the real bottleneck in building a system.
A worked example
You train a model on emails each marked spam or not spam, and it learns to predict the label for emails it has never seen.
Common confusion
Supervised learning needs labelled examples. Unsupervised learning, by contrast, finds patterns in data that has no labels at all.