The two basic shapes of prediction problem. Classification puts things into categories: spam or not spam, fraud or legitimate, which of five faults this is. Regression predicts a number: tomorrow’s demand, the price of a house, how long a repair will take. Almost every practical machine learning task is one or the other, and knowing which you have decides how you measure success.
Technical view
Classification outputs a distribution over discrete classes, typically trained with cross-entropy loss and evaluated with accuracy, precision, recall and the area under the receiver operating characteristic curve. Regression outputs a continuous value, typically trained and evaluated with mean squared or mean absolute error. The boundary blurs in practice: ordinal problems and probability estimation sit between the two, and a classifier’s raw scores usually need calibration before they can honestly be treated as probabilities.
Related
Part of AI Terminology Index, section: Classical Machine Learning.