SVM Cross Validation Training

I am using K-Fold cross validation. My K is 10.

I am supposed to do 10 crossfold and take the average of the SVM performance.

How should i perform such? Running the cross validarion ounce only generates 1 fold prediction or a complete 10-fold prediction?

[Mohammad Sami]

According to the documentation it is average over all folds

https://www.mathworks.com/help/releases/R2020a/stats/select-data-and-validation-for-classification-problem.html

Answer 1

[Gayathri]

Hi @Nedz,

I understand that you need to perform K-fold cross-validation for a SVM model. For this purpose you can use the "crossval" function. And then, "kfoldLoss" function can be used to get the classification loss for cross-validated classification model. Please refer to the code below which implements the same.

load ionosphere
%Train a SVM classifier using the radial basis kernel
SVMModel = fitcsvm(X,Y,'Standardize',true,'KernelFunction','RBF','KernelScale','auto');
%Cross-validate the SVM classifier
CVSVMModel = crossval(SVMModel);
%Estimate the out-of-sample misclassification rate.
classLoss = kfoldLoss(CVSVMModel)

"crossval" by default uses 10-fold cross-validation.

Please refer to the "Train and Cross-Validate SVM Classifier" example in the documentation link mentioned below.

https://www.mathworks.com/help/stats/fitcsvm.html

Hope you find this information helpful!