Compare Leave-One-Out cross-validated performance of classifier across two datasets I am trying to determine whether there is a statistically significant difference between the performance of a SVM classifier on two different datasets. The performance of the classifier is determined through Leave-One-Out cross-validation. Each dataset has 480 samples, and is binary categorical: correct prediction or incorrect prediction.
It is my understanding that it isn't possible to use a chi-squared test as the separate outcomes from cross-validation tests are not independent within  the dataset itself - the same sample is used 479 times. 
If anybody could suggest to me where to look or how to go about this, I would really appreciate it!
 A: The following paper is one of the main literature that discussed statistical significance tests.
Dietterich, T. G. (1998). Approximate statistical tests for comparing supervised classification learning algorithms. Neural Computation, 10(7), 1895–1923. 
https://sci2s.ugr.es/keel/pdf/algorithm/articulo/dietterich1998.pdf
A: The paper suggested by @Fatima is very interesting and proposes multiple tests to pick the best model on a dataset. 
Given that instead in this problem you have that you want to find the best dataset given the model, the approach should be slightly different because we are not in a "paired observations" test case, since the two datasets are different and uncorrelated. 
I would definitely avoid using Leave One Out CV, as it is extremely expensive computationally. What you can do is get results via a simple 5x2 Cross Validation (or even more, like 10x2 if you have the time). These results will be uncorrelated between the two datasets, and slightly correlated between them (in particular, they will be correlated in pairs).  You can then run a simple t-test for the means of the two populations (on dataset 1 and dataset 2) to see if there is any difference. If you want to reduce correlation further (better for the test), you can keep only one of the two results from every CV.
