What aspect of data is preserved in stratified k-fold cross-validation?

Stratified k-fold cross-validation is a technique used in the validation of machine learning models that maintains the proportion of different classes in each fold of the dataset. This is particularly important in classification problems where there may be an imbalance in class distributions. By ensuring that each fold reflects the original dataset's class distribution, stratified k-fold cross-validation helps to provide a more accurate estimate of the model's performance on unseen data. This approach minimizes the risk of the model being biased toward the majority class, which can occur if folds are created randomly and do not maintain class proportions. In doing so, it preserves the underlying data distribution, making it more robust for performance evaluation.