Cross-validation
Also called K-Fold Cross-Validation
Rotate which slice of data is held out so every example gets to be both study material and test, then average the scores.
Think of it like
Five friends take turns being the quiz-master while the others answer, so everyone gets tested and no single split decides it.
Example
With only 2,000 samples, a team runs 5-fold cross-validation and reports the average accuracy across the five folds.
How it actually works
Split the data into k folds; train on k−1 and evaluate on the one left out, cycling through so each fold is the holdout once. Averaging the k scores gives a more stable estimate than a single split, which matters most when data is limited. The cost is training k times, so it is common in classic ML and rare for giant models.
For product teams
Buys you a trustworthy number when you do not have enough data to spare a big test set.
For engineers
Partition into k folds, train/evaluate k times rotating the held-out fold, and average the metric.
Related
- Holdout — The single-split version it generalizes.
- Validation set — A per-run holdout used during tuning.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome