I am trying to use gbm for a classification problem and my code to calculate the roc curve is:
pred.gbm <- predict(fit.gbm, newdata=test.data[,1:18], n.trees=best.iter, type="response") pred.label <- 1*(pred.gbm>0.103) #1: >0.103; 0: otherwise Confusion table table(pred.label,test.data[,20]) gbm.roc.area(test.data[,20],pred.label)
The confusion matrix is:
However the area is coming out to be around 0.50.I think this has to do with the probabilities to generate pred.label.
Is there a way to calculate a graph of specificity vs sensitivity to see the optimum cutoff in gbm??