What is the name of the cross-validation method that splits a dataset into training and test sets?

The cross-validation method that specifically involves splitting a dataset into distinct training and test sets is known as the holdout method. This approach is straightforward and commonly used in machine learning. It involves randomly dividing the dataset into two subsets: one for training the model and another for evaluating its performance. By doing so, it allows for a clear assessment of how well the model generalizes to unseen data.

In the holdout method, typically a predetermined proportion of the data is allocated to the training set (often around 70-80%) and the remainder to the test set. This direct split ensures that the evaluation metrics derived from the test set provide an unbiased estimate of the model’s performance.

Other cross-validation methods, such as leave-one-out or k-fold, involve more complex partitioning strategies that repeatedly train and test models on various subsets of the data for robustness in performance measurement. Stratified cross-validation is specifically designed for maintaining the distribution of classes in the fold splits, which is particularly useful in imbalanced datasets, but it still relies on the general concept of repeated splitting rather than a single holdout evaluation.