A data scientist is building a binary disease classifier from a random sample of 400 patients, where the disease prevalence is about 3%. Which cross-validation strategy is most appropriate to evaluate the model?

Stratified k-fold cross-validation is most appropriate because the dataset has a highly imbalanced class distribution (3% disease prevalence). Stratification ensures that each fold maintains approximately the same proportion of positive and negative samples as the overall dataset. This prevents folds from having too few or no positive samples, which can happen with standard k-fold cross-validation and lead to unreliable model evaluation. Standard k-fold cross-validation with or without repeats would not guarantee balanced class distribution per fold. An 80/20 train/validation split, while useful, is a single split and doesn't offer the robust evaluation of multiple folds provided by cross-validation.