15

I have a dataframe. first column contains my the predictive score (range from 0 to 100, smaller values is expected to be in class A, larger values is expected to be in class B) for my model, 2nd column contains the real classification of the entries (either "class A" or "class B").

How to get confusion matrix with R for different cut off values, as I cannot decide where I should define values < 20 or < 50 as class A yet?

How to do this comparison efficiently with R?

vhadalgi
  • 7,027
  • 6
  • 38
  • 67
user2718
  • 449
  • 1
  • 7
  • 13

1 Answers1

22

There's a number of ways to do this, a reproducible example with your data would have been desirable:

set.seed(12345)
test <- data.frame(pred=c(runif(50,0,75),runif(50,25,100)), group=c(rep("A",50), rep("B",50)) )
table(test$pred<50,test$group)

gives

         A  B
  FALSE 18 34
  TRUE  32 16

So this says 32 A's were under 50 and 34 B's were over 50, while 18 A's were over 50 (wrongly classified) and 16 B's were under 50 (wrongly classified)

set.seed(12345)
test <- data.frame(pred=c(runif(50,0,60),runif(50,40,100)), group=c(rep("A",50), rep("B",50)) )
table(test$pred<50,test$group)

gives

         A  B
  FALSE  8 40
  TRUE  42 10

In this example, cause of the chosen sampling, your classification is much better.

The '50' in this can then be changed to anything you want, 20, 30, etc.

table(test$pred<50,test$group)
nzcoops
  • 9,132
  • 8
  • 41
  • 52
  • 1
    This does not work if you do not have predictions lower than 50, since the table only produces a TRUE row and not a 4*4 matrix. – Zhubarb Sep 11 '14 at 13:13
  • I edited the answer to generalise. * (I meant 2*2 matrix in the above comment). – Zhubarb Sep 11 '14 at 14:06