What Is Cross-Validation? A Plain-Language Explanation
Cross-validation is a resampling technique that splits a dataset into several groups, repeatedly trains the model on some groups and tests it on the remaining group, then averages the results to estimate how well the model will generalize to unseen data.
The simple analogy
Rather than judging a student from a single test that might happen to suit them, you give them several tests covering different material and average the results. Cross-validation rotates which part of the data is held out so the estimate depends less on one lucky split.
Why a single split is risky
A single train-test split can give an optimistic or pessimistic result depending on how the data happened to be divided. Cross-validation uses every example for both training and validation across rounds.
How k-fold works
| Step | Description |
|---|---|
| Split | Divide data into k folds |
| Rotate | Use one fold as the test set |
| Train | Fit on the other k-1 folds |
| Repeat | Cycle through all k folds |
| Average | Combine the k scores |
Common choices
A value of k = 5 or k = 10 is commonly used, balancing reliability and computation time.
Honest caveats
Cross-validation trains the model multiple times, which increases computation. It also helps detect overfitting but does not replace careful data preparation or a final evaluation on a separate test set.
FAQ
What is a fold?
One of the subsets the data is divided into, used once as the validation set.
What does k mean?
The number of folds, which equals the number of training rounds.
Does cross-validation prevent overfitting?
It helps detect overfitting and guides model selection, rather than preventing it directly.
Why average the scores?
Averaging reduces the influence of any single split and gives a more stable estimate.
Key takeaways
- Cross-validation rotates held-out folds across multiple rounds.
- It estimates generalization more reliably than one split.
- k = 5 or 10 is common.
- It costs more computation.
Continue with what is a confusion matrix and what is model evaluation.
Sources
Learn by doing
Read a guide, then practice with bite-sized AI lessons you can take offline.
Start learning free