Cross-Validation

What Cross-Validation is

Rather than trusting a single train-test split, cross-validation rotates the held-out fold so every record is evaluated once, giving a mean and a variance for the score.

How it works

k-fold splits the data into k parts; stratified variants preserve class balance; time-series validation only ever tests on later periods. All preprocessing must be fitted inside each fold.

Why it matters

It makes small-data results trustworthy and exposes unstable models that a lucky split would hide.

Common uses

• →Hyperparameter search
• →Model comparison
• →Small dataset evaluation