Glossary · Phase 3 · Machine learning
Undersampling
Training on every example of the rare class and a random sample of the common one, so the classes are balanced; it throws data away.
Try it now
Slide the threshold and watch precision and recall trade places. Then lower the share who churned and see accuracy stop telling you anything.
Learn it properly
Precision, recall and F1 · Accuracy lies; these don't. About 30 minutes.
Other terms from the same lesson
- class weights: Making each example of a rare class count for more in the loss, so a model pays attention to it without discarding any data.
- F1 score: One number that balances precision and recall.
- false negative: A no the model got wrong: it predicted the customer would stay, but they churned.
- false positive: A yes the model got wrong: it predicted churn, but the customer stayed.
- harmonic mean: An average that stays low if either number is low. F1 is the harmonic mean of precision and recall.
- macro average: Averaging a score across classes with each class counted equally, however rare it is.
- precision: Of the things a model flagged, the share that were right.
- recall: Of the things that should have been flagged, the share the model caught.
- true negative: A no the model got right: it predicted the customer would stay, and they did.
- true positive: A yes the model got right: it predicted churn, and the customer did churn.
Never trained an AI?
Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account.