Cross-Validation Quiz

Answers and explanations appear as you take the quiz.

1. In standard k-fold cross-validation, how many times is each data point used for testing?
   • Exactly once
   • k times
   • k-1 times
   • It depends on the random seed
2. What is the main advantage of k-fold cross-validation over a single train/test split?
   • It always increases model accuracy
   • It gives a more stable, less luck-dependent estimate of generalization
   • It removes the need for a test set entirely
   • It speeds up model training
3. Stratified k-fold cross-validation is especially useful when:
   • The dataset is perfectly balanced
   • The data is a time series
   • You have imbalanced classes and want each fold to keep similar class proportions
   • You want to train on the entire dataset at once
4. Leave-one-out cross-validation (LOOCV) sets the number of folds k equal to:
   • 2
   • 10
   • The number of features
   • The number of samples in the dataset
5. To avoid data leakage during cross-validation, preprocessing steps like feature scaling should be:
   • Fit only inside each training fold, then applied to the validation fold
   • Fit on the full dataset before splitting into folds
   • Skipped entirely
   • Done only on the validation folds
6. Why is ordinary k-fold cross-validation inappropriate for time-series forecasting?
   • Time series have too few rows
   • Random folds can put future data in the training set used to predict the past
   • It is too fast for time series
   • Time series cannot be split at all
7. What is the purpose of nested cross-validation?
   • It makes every model use the same hyperparameters
   • It lets preprocessing use the full dataset before splitting
   • It separates hyperparameter selection from an outer evaluation of the selected procedure
   • It removes the need for held-out evaluation
8. Which fold counts are common starting choices for k-fold cross-validation?
   • Exactly two folds in every problem
   • One fold regardless of sample size
   • One thousand folds for every dataset
   • Five or ten folds