Learning from labelled examples.
WHAT IT IS
Training on examples where the correct answer is provided.
Photographs labelled cat or dog. Emails labelled spam or not. Loans labelled repaid or defaulted.
HOW IT WORKS
The system predicts, compares its prediction with the label, adjusts, and repeats across the dataset many times.
WHAT IT PRODUCES
A model that predicts labels for new, unlabelled input.
THE TWO MAIN TYPES
Classification: predicting a category. Spam or not. Which of five products.
Regression: predicting a number. A price, a temperature, a duration.
WHAT IT REQUIRES
Labelled data, in quantity. That is usually the constraint.
Labelling is expensive and frequently the largest cost of a project.
WHERE IT IS USED
Fraud detection Medical image screening Credit scoring Demand forecasting Content moderation
THE LIMITATION
It learns the patterns in the labels it was given, including any bias in how they were assigned.
