Always choosing Repeated CV instead of fold CV To my understanding, Repeated CV gives superior results compared to k fold CV.
Is there any case where k fold CV is preferable compared to Repated CV?
In particular for Neural Networks, what would be the preferable choice?
I use R and caret package
Regards,
Ioannis
 A: Repetition and folding are orthogonal concepts. Folding means splitting your data into k similar subsets and using all combinations of k-1 as training and using the remaining fold as testing. Repetition is repeating whatever cross validation procedure you used many times (provided the internal random number generator generates different partitions/subsets - only be sure not to use the same seed for each repetition). 
What you call repeated CV is probably a repeated x% hold-out, where you use x% of the data as test, and the other 100-x% as training. 
AS FAR AS I KNOW, from estimation of a parameter (accuracy, error rate, and so on) there is no difference between folding and repeated hold-out PROVIDED you use the same number of training/testing sets and they are of the same size. Thus a 5-fold estimation is as good as a 5x repeated 20% hold-out estimation. A 2x repeated 3-fold is equivalent to a 6x repeated 33% hold-out.
The advantage of a repeated hold-out is that you have more control: you can have a 6x repeated 10% hold-out. But a 10-fold would repeat the estimation 10 times.
