AUC for 3-Class prediction in R

Question

I am building a multinomial logistic regression on top of the iris datas-set, to segregate it into 3 classes.

I'm using the mulitnom() function from the nnet package:

library(nnet)

mod1 = multinom(Species~., data = iris, trace = F)

After building the model, i am predicting the results on top of the iris data-set:

predict(mod1,iris)->result1

table(iris$Species,result1)

This only gives me a confusion matrix, but i want to find the AUC for these three classes, how can i do that?

Answer 1

• Use type='probs' inside predict() function, so that you get the probabilities.

result1 = predict(model, data = iris, type='probs')

• Then you can use the auc-multcap method from the handtill2001 package:

library(HandTill2001)
auc(multcap(
  response = iris$Species,
  predicted = as.matrix(result1)
))

This gives you an AUC value of  0.9990667