K Fold Split Calculator

Calculator Inputs

Total Samples

Number of Folds (k)

Starting Sample Index

Holdout Test Split Mode

Test Percentage

Fixed Test Samples

Shuffle Samples

Random Seed

Remainder Placement

Reset

Example Data Table

This example uses 120 total samples, 10% holdout testing, and 5 folds.

Fold	Validation Samples	Training Samples	Cross Validation Pool	Holdout Test Samples
1	22	86	108	12
2	22	86	108	12
3	22	86	108	12
4	21	87	108	12
5	21	87	108	12

Because 108 ÷ 5 = 21 remainder 3, three folds receive one extra validation sample.

Formula Used

1) Holdout test set: If you choose a test split, the calculator removes those samples first.

CV Samples = Total Samples − Test Samples

2) Base validation size: Each fold receives the integer part of the remaining pool divided by k.

Base Fold Size = floor(CV Samples / k)

3) Leftover samples: Any remainder is distributed to early or late folds, depending on your selected placement rule.

Remainder = CV Samples mod k

4) Training size for each fold: The training set uses every cross validation sample except the current fold’s validation block.

Training Samples per Fold = CV Samples − Validation Samples in Fold

5) Reuse count: Every cross validation sample appears in training exactly k − 1 times across the full cycle.

How to Use This Calculator

Enter the total number of dataset samples.
Choose how many folds you want for cross validation.
Optionally reserve a holdout test set by percentage or fixed count.
Set whether samples should be shuffled before splitting.
Provide a seed when you want repeatable fold assignments.
Choose the starting index if your sample numbering starts from 0, 1, or another value.
Select whether remainder samples should go to early or late folds.
Press Calculate K Fold Split to view summary metrics and per-fold details.

FAQs

1) What does k fold splitting do?

It divides a dataset into k validation groups. Each group is used once for validation, while the remaining groups are used for training.

2) Why would I keep a separate test set?

A holdout test set gives you one final untouched benchmark after cross validation finishes. It helps reveal whether tuning choices generalized well.

3) How should I choose the value of k?

Common choices are 5 or 10. Smaller values train faster, while larger values often reduce bias but increase runtime and repeated fitting cost.

4) When should I enable shuffling?

Enable shuffling when sample order may contain patterns, such as sorted classes, timestamps, or grouped observations. It usually creates more balanced folds.

5) Why is a random seed useful?

A seed makes shuffled results repeatable. That helps debugging, reporting, and collaboration because everyone can regenerate the same folds later.

6) What happens when samples are not divisible by k?

The leftover samples become the remainder. This calculator places them in early or late folds, depending on your selected distribution rule.

7) Does this calculator perform stratified splitting?

No. It computes standard fold sizes and sample assignment previews. Stratified splitting needs class distribution data for every sample.

8) Can I use this for very small datasets?

Yes, but use care. If the dataset is tiny, each fold becomes unstable, and model estimates may vary sharply across validations.