Cross-Validation Calculator

Fold Size is Dataset Size divided by K-Folds, so more folds means each individual fold is smaller. Training Size per Fold is what's left over after removing one fold (Dataset Size minus Fold Size), so as Fold Size shrinks with more folds, Training Size per Fold grows correspondingly closer to the full dataset -- the two outputs move in opposite directions because they're two halves of the same split.

Data Utilization is the percentage of your full dataset that ends up in the training set for any single fold -- Training Size per Fold divided by Dataset Size. It rises toward (but never reaches) 100% as K-Folds increases, since a larger K means each fold holds out a smaller slice for validation and trains on a correspondingly larger share of the remaining data.

The Test Set is held out entirely and used only once, at the very end, to report a final unbiased performance figure -- it should never influence any training or tuning decision. The Validation Set is carved out of what remains after the test split and is used during development (for example, to tune hyperparameters or decide when to stop training), which keeps those tuning decisions from leaking information into the untouched Test Set.

More folds (like 10) train each model on a larger share of the data per fold, which this calculator's Data Utilization figure reflects directly, but it also means more separate training runs and a smaller, noisier held-out fold each time. Fewer folds (like 3) train faster and average over larger held-out chunks but use less data per training run. 5 and 10 are the most common defaults in practice as a balance between the two, though this calculator doesn't pick one for you -- it just shows you the resulting split sizes.