Classification
A supervised task that assigns each input to one of a discrete set of categories.
- Binary (spam/ham), multiclass (digit 0-9), multilabel (multiple tags).
- Output is typically a per-class probability plus arg-max.
- Picking the right metric (accuracy, F1, AUC) starts here.
When to use: When the target is a category, not a number: churn, click, default, convert, intent.
Example: Email spam filter: input is text + headers, output is {spam, not-spam}. Trained on millions of labeled emails, serves at sub-millisecond latency.