Machine learning (ML) is the part of AI in which a model learns a pattern from examples. You do not write a separate rule for every case. You show the model data, and it estimates a function that works on new cases.
The three main families
| Family | What you provide | What the model learns | Example |
|---|---|---|---|
| Supervised | Inputs and the correct answer | How to predict that answer | Will this customer churn? |
| Unsupervised | Inputs only | Groups or structure in the data | Which customers behave alike? |
| Reinforcement | Actions and a reward | A policy that earns a higher reward | Which offer to show next |
Features and labels
A feature is an input the model can use, such as income, tenure, or number of late payments. A label is the outcome you want to predict, such as defaulted or not. In supervised learning you train on both. At prediction time you have features only, and the model estimates the label.
A simple pipeline
- Define the decision and the label.
- Collect historical examples and clean them.
- Split the data so the test set is unseen.
- Train a model and measure it on the test set.
- Deploy it, then watch whether the data or the accuracy drifts.
A practical example
A lender wants to estimate default risk. Features include income, years in the job, existing debt, and repayment history. The label is whether a past applicant defaulted within 12 months. The trained model scores a new applicant. A credit officer still applies policy; the model supplies a consistent signal.
What to remember
- ML learns from examples. It is not a fixed script of rules.
- The label must match the decision you will actually make.
- A model that looks strong on old data can fail when the world changes. Measure it on fresh cases.
