Classification

What Classification is

Classification covers binary decisions such as fraud or not fraud, and multi-class decisions such as routing a support ticket to one of twenty queues.

How it works

A model outputs a score per class, usually normalised into probabilities. A threshold or argmax converts that into a decision, and the trade-off between precision and recall is tuned to match the cost of each error type.

Why it matters

It is the most common shape of applied machine learning problem in business, and its evaluation metrics are the ones non-specialists most often misread.

Common uses

• →Spam detection
• →Ticket routing
• →Image labelling
• →Medical triage