K-fold cross validation with more folds in Maxent

Question

Hi I'm using the Maxent software 3.4.0 for Mac and I'm trying to understand an issue about k-fold Cross-validation.

Basically, I understood that my dataset is splitted in k folds and each fold more or less has the same size. Therefore, if my dataset has 100 observations, a 10-fold cross validation will split the dataset in 10 folds of 10 observations, and Maxent will train 10 models, each with 9 folds, and the 10th will test it.

My question is: can I split my dataset in more than 10 folds (e.g. 50 folds), BUT with 10 observations per fold? In this case of course occurences would not be used once, but as many times as they appear in the different folds.

Can I do it (without the command line, that I don't know how to use it)? Could be the result meaningful?

Answer 1

The point of cross-validation is that each iteration your model is tested on observations that it has not been calibrated on. In your example, it will be inevitable that your validation folds will contain observations used in model calibration, inflating the cross-validated AUC.

What you could look at is using the bootstrapping option in Maxent. A question on cross validation and bootstrapping with Maxent was asked previously here FYI https://gis.stackexchange.com/questions/366513/difference-between-bootstrap-and-cross-validation-maxent