Cross-Validation Strategies

Advanced Cross-Validation

Different data require different validation strategies.

Stratified K-Fold

StratifiedKFold(n_splits=5) maintains class proportions. Essential for imbalanced classification.

RepeatedStratifiedKFold repeats for stability.

Group K-Fold

GroupKFold(n_splits) ensures groups stay together. Use when data has group structure.

Prevents data leakage across groups.

Time Series Split

TimeSeriesSplit for temporal data. Each fold uses earlier data for training, later for validation.

Key Takeaways

1. Stratified K-Fold maintains class balance
2. Group K-Fold prevents leakage across groups
3. Time series split respects temporal order