A technique that splits data into multiple subsets to train and test models, reducing overfitting risk.