Cross-Validation

Updated 17 Sep 2026

•
What it is
- •
  evaluating a learning procedure across repeated training/validation partitions
•
k-fold procedure
- •
  divide data into $k$ folds
- •
  train on $k-1$ folds and evaluate on the held-out fold
- •
  repeat so each fold is held out once
•
Variants
- •
  stratified folds approximately preserve class proportions
- •
  grouped folds keep related observations together
- •
  time-aware splits preserve temporal ordering
•
Caution
- •
  stratification does not remove dependence or solve class imbalance by itself
- •
  preprocessing must be fitted separately inside each training fold
- •
  when cross-validation is used for tuning, an independent test set or nested evaluation is needed for an appropriate final performance estimate
- •
  the handout’s blanket instruction to use stratified k-fold for limited or imbalanced data is too broad; the sampling structure determines the appropriate splitter.
•
Related
- •
  Dataset Splitting

Linked references 2