What do the values from varImp() for a randomForest indicate?




What do the values from the varImp() function of the caret package for a randomForest model indicate? For one of my models, it shows negative values for some variables and positive for some. What is the significance of both of these?



hi @pravin,

The first plot basically measures how badly the model performs without each variable,so for highly predictive variables,the decrease is expected to be significant-like Title

The second plot digs into the maths behind decision trees but basically computes the importance of the variable in the model.
However I have used the randomForest package in R and this plot can be gotten by:


Thanks for the explanation but I was asking about the result of varImp() and not varImpPlot(). And also what is the relation between the output of these two.