k-Fold Cross-Validation
Definition
In k-fold cross-validation, a dataset is divided into k subsets and the model is evaluated k times with changing validation sets.
In each run, k-1 folds are used for training and one fold for validation. The results are aggregated to obtain more stable estimates than with a single random split.
Simply explained
Like performing the same test multiple times, each time withholding a different part of the tasks.
Why it matters in practice
Relevant for model comparison with limited data, as long as data dependencies and temporal structure are considered.
Example from business
Several classifiers are compared using 5-fold CV before a final candidate is tested on a separate test set.
Limits and pitfalls
- Including the test set in CV; randomly shuffling time series; leaking grouped data across folds; reporting only the mean without variance.
Origin & history
Cross-validation is an established statistical resampling method. Kohavi empirically compared various procedures for classification and model selection in 1995.
Use in mid-sized companies
- Possible applications: model selection, hyperparameter comparison, and robust estimation with smaller datasets.
Frequently asked questions
Does CV replace a final test set?
Not necessarily. Is k=10 always optimal? No, effort and data structure determine a sensible choice.
Sources
- A Study of Cross-Validation and Bootstrap · Kohavi (1995), A Study of Cross-Validation and Bootstrap
Sources checked on 10.10.2026.
Editorially reviewed on: