Caret Package: Stratified Cross Validation in Train Function

Question

Is there a way to perform stratified cross validation when using the train function to fit a model to a large imbalanced data set? I know straight forward k fold cross validation is possible but my categories are highly unbalanced. I've seen discussion about this topic but no real definitive answer.

Thanks in advance.

I'm also looking for the answer... By default, function createFolds() creates stratified folds. But I'm not sure about the train function when using method = "cv" in trainControl. — jbrettas, Apr 21 '16 at 20:38

score 17 · Accepted Answer · answered Sep 19 '16 at 02:52

17

There is a parameter called 'index' which can let user specified the index to do cross validation.

folds <- 4
cvIndex <- createFolds(factor(training$Y), folds, returnTrain = T)
tc <- trainControl(index = cvIndex,
               method = 'cv', 
               number = folds)

rfFit <- train(Y ~ ., data = training, 
            method = "rf", 
            trControl = tc,
            maximize = TRUE,
            verbose = FALSE, ntree = 1000)

answered Sep 19 '16 at 02:52

KST

589
9
15

This is stratified by a single factor. There doesn't seem to be an option for multi-factor. – jiggunjer Jan 15 '20 at 08:16

Caret Package: Stratified Cross Validation in Train Function

1 Answers1

Linked