Neural Network training using LeaveMout cross-validation

I have data for 8 different environmental variables (Input.mat ; please see attachment), using which I want to predict a new parameter (Target.mat ; please see attachment). I have Input and Training data for 22 dates. I need a Neural Network (NN) through which I can predict the values of the targets. As I don't have much data for dividing it into three parts (i.e. Training, Validation and Test), so I need a NN to use the a LeaveMout cross validation method to give me the best Neural Network for the prediction of the desired targets. Currently, I have been using the attached code (NN_Environ.m ; please see attachment) for the NN training but I am not sure how can I introduce the "LeaveMout" cross validation method for the said problem. I don't know how to code it in MATLAB, your help will be appreciated. Any suggestions regarding the usage of other cross-validation method which could work best for my problem are welcome.

A special request to Greg Heath for his valuable suggestions on this.

Thanks and Regards, Majid

Answer 1

[Greg Heath]

Since there is no NNToolbox code for M-fold cross-validation, it is just easier to make multiple designs with random datadivision and weight initialization because these are defaults.

In zillions of examples I have used double-loop configurations with the outer loop over a number of hidden nodes values (h = Hmin:dH:Hmax) and the inner loop over Ntrials (i=1:Ntrials) random divisions and initializations.

If for, some reason, you have to use M-fold, I do have a few examples in the NEWSGROUP and ANSWERS. Search

 greg k-fold      11 NEWSGROUP HITS + 26 ANSWERS HITS

Hope this helps.

Thank you for formally accepting my answer

Greg