Definition
Supervised learning is teaching software to predict answers by training it on past examples where the correct answer is already labeled.[1]
At a glance
- It learns from labeled examples, data tagged with the right answer (this email is spam, this loan defaulted).[1]
- Two main jobs: classification (pick a category) and regression (predict a number like price or demand).[3]
- The payoff is prediction on new, unseen data, flagging fraud or forecasting sales automatically.[2]
- It is only as good as your labels: messy or biased examples produce messy or biased predictions.
How it works in plain terms
You feed the system many past records where the outcome is known, say thousands of transactions marked fraud or legitimate. It studies the patterns linking the inputs to those outcomes.[4] Afterward it can score a brand-new transaction and predict whether it is likely fraudulent, no human reviewing each one.
Where businesses use it
Spam filters, fraud detection, credit-risk scoring, customer-churn prediction, demand forecasting, and disease screening from medical images all run on supervised learning.[3] The common thread: you have historical data with known results and want the same kind of answer on future cases at scale.
Bottom line
If you have past examples with known right answers, supervised learning turns them into a tool that predicts those answers on new cases automatically.