1
$\begingroup$

I'am working on a 4 class Brain computer interface and i want to validate my work , in so many academic papers they write 10x10 fold cross validation

What i want to know what is the 10x10 fold cross validation And is it different from 10 fold cross validation

$\endgroup$
5
$\begingroup$

Cross-validation can be done once (10 times, or simply 10), or done ten times (10 times done 10 times, or 10 x 10). For clarity, let's change these values to 5 and 5 x 20. Here's how this works:

  1. The training set is randomly partitioned into five chunks.
  2. Five models are built out of these five. Each model is built using four of the chunks as a training set, and one as a testing set.
  3. The five models are averaged to create a single model balanced for randomness.
  4. To account for the initial choice of the five partitions, we repeat steps one through three 20 times. The resulting 20 models are averaged.

Steps 1-3 constitute cross validation, which is the 10 you're asking about, or the 5 in the example. Step 4, which repeats steps 1-3, is repeated portion of repeated cross-validation and the x 10 you're asking about, or the x 20 in the example.

$\endgroup$
  • 2
    $\begingroup$ Note that in my field (chemometrics) 5 x 20-fold cross validation would mean 20-fold cross validation (i.e. partition into 20 chunks), repeated 5 times. => conclusion: always spell out what is meant. $\endgroup$ – cbeleites supports Monica Jul 17 '17 at 10:49
  • $\begingroup$ The above tends to be more common, though it is rarely clarified. Thank you for pointing this out. In addition to the above ambiguity, there is also some disagreement on how the folds should be treated in per-fold models. In the answer, I suggested that "each model is built using four of the chunks as a training set, and one as a testing set." Another approach is to use one chunk for training and the rest for testing. This should produce worse estimates and more variance between the folds, but sets a more rigorous requirement for consistency and is computationally cheaper. $\endgroup$ – Alex Firsov Jun 4 '19 at 22:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.