← AI glossary

Definition

Supervised Learning

Supervised learning trains models on labeled examples so they can predict labels, scores or values for new data.

Also known as: labeled learning

Short definition

Supervised learning is a machine learning setup where each training example includes both input data and the correct answer. The model learns to map inputs to outputs and is later tested on new cases.

How it works

The training data might pair customer records with churn labels, images with object labels or transactions with fraud outcomes. The model makes predictions, compares them to labels and updates itself to reduce mistakes.

Example

A bank can train a supervised model on past transactions marked as fraudulent or legitimate. When a new transaction arrives, the model estimates the probability that it is suspicious.

Why it matters

Supervised learning is practical when good labels exist. Its quality depends heavily on label accuracy, dataset coverage and evaluation. If historical labels contain bias or errors, the model can learn and repeat those problems.

Train, validation and test data

Examples are normally divided into training, validation and test sets. The model learns from the first, design choices are made with the second and final performance is measured on the third. Test data must not leak into training, or the reported result will be artificially optimistic.

Choosing the right metric

Accuracy alone can be misleading. If fraud appears in one transaction out of a thousand, a model that always predicts “legitimate” looks accurate while being useless. Teams may need precision, recall and explicit estimates of the cost of a false alarm versus the cost of a missed case.