Cross-Validation

K-Fold, stratified, LOO, and train/val/test split — fold diagram, bias-variance tradeoff of k, CV score aggregation with CI.

Dataset

Dataset size NNumber of folds k (2–20)

Method

Fold Diagram

■ validation■ training

Bias-Variance Tradeoff of k

K-Fold Summary

N500

k folds5

Train size / fold400

Val size / fold100

Data efficiency0.8

Recommendationk=5

N < 1000: k=5 balances bias and variance

CV Score Aggregation

Validation scores (one per fold)

Train/fold = ⌊N/k⌋ × (k−1) · Efficiency = (k−1)/k