What is cross-validation and when should you use different types?

Cross-validation estimates model performance by training and testing on different data subsets to reduce overfitting to specific train-test splits.\n\n• K-fold: Standard approach, good for most problems\n• Stratified: Maintains class distribution, essential for imbalanced data\n• Time-series: Respects temporal order, prevents data leakage\n• Leave-one-out: Maximum data usage, computationally expensive\n\nExample: Use stratified 5-fold CV for classification, time-series split for financial data, nested CV for hyperparameter tuning. Always ensure no data leakage between folds.