How many models does repeated k-fold CV train?

Question

You run repeated 5-fold cross-validation: the entire 5-fold procedure is carried out 4 times, each time with a different random split of the data into folds, to average out the luck of any single split. How many separate models are trained in total?

Accepted Answer

A single pass of 5-fold CV trains 5 models (one per fold). Repeating the whole thing 4 times with fresh random splits multiplies that: 5 	imes 4 = 20 	ext{ models}. Repeated CV reduces the variance of the performance estimate that comes from an unlucky fold assignment, at the cost of extra fits. Here it is 20 fits instead of 5. Takeaway. Repeated k-fold CV with r repeats trains k 	imes r models; here 5 	imes 4 = 20.

How many models does repeated k-fold CV train?

Your answer

Discussion

Learn the concepts

Related questions