Spam or not spam. Malignant or benign. Fraud or legitimate. Classification assigns inputs to one of several predefined categories, and it underpins a huge share of practical machine learning.

Models learn a decision boundary from labeled examples. Binary classification separates two classes. Multiclass extends to several. Multilabel allows multiple simultaneous labels, useful when categories are not mutually exclusive.

Common classifiers

• Logistic regression
• Decision trees and random forests
• Support vector machines
• Neural networks
• Naive Bayes

Accuracy alone is a poor metric when classes are imbalanced. A fraud detector that flags nothing achieves 99.9 percent accuracy on a dataset where fraud is rare, yet it is useless. Precision, recall, and F1 score tell a more honest story.