Classification is used when the output is a category rather than a number.
An email is classified as spam or not spam.
A medical scan is sorted into one of several diagnosis categories.
The model learns boundaries that separate one class from another.
Logistic regression, decision trees and neural networks can all classify.
Classification means the answer is one of a fixed set of choices.
Spam or not spam, cat or dog, approve or decline.
Most classifiers do not hand you a class straight away.
They give a score for each class, and something turns that into a decision.
For two classes that something is usually a threshold like 0.5.
Moving the threshold trades one kind of mistake for another.
Lower it and you catch more real cases, but you also raise more false alarms.
Which mistake matters more is a business question, not a maths one.
The default threshold is a habit, not a law.
Class balance changes everything about how you judge the model.
With ninety-nine rare cases in a million, always saying no is right almost every time.
So accuracy is often the worst measure you could reach for.
More than two classes brings its own trouble.
The model can be excellent on the big classes and useless on the small ones.
Always look at the mistakes class by class, not just the overall figure.
One question on this card. Every option explains itself once you pick, right or wrong.
55 concepts in this module, then a graded test with every answer explained.