K-fold cross validation partitions the dataset into k equally sized folds, iteratively using one fold for testing and the remaining folds for training. By rotating through all folds, the model is evaluated on multiple, non-overlapping subsets of data, providing a robust estimate of generalization performance. This process reduces the risk of overfitting to a single train–test split and is explicitly recognized in AI testing techniques and data analytics practices relevant to AAIA.

Option A (failure mode effects analysis) is useful for understanding failure impacts but not primarily for generalization. Option B (unit testing) focuses on code components, not model generalization. Option D (metamorphic testing) is valuable for validating behavior under transformations, but the standard, widely accepted approach to assessing generalization remains cross validation , making option C the best answer.