Cross-Validation

A single train/test split gives a high-variance estimate of generalization performance — it depends heavily on which samples ended up in each split. Cross-validation averages over K different splits for a more reliable estimate.

1. Split data into K equal folds 2. For each fold i: train on all folds except i, evaluate on fold i 3. Average the K evaluation scores K=5 or K=10 is standard. K=N (Leave-One-Out) is most thorough but slow.

Preserves the class distribution in each fold. Always use this for classification tasks — especially with imbalanced data.

For time-series data, you must never let future data leak into training. TimeSeriesSplit always trains on past data and validates on future data.