user697911 user697911 - 2 months ago 9
R Question

How to retrive overall accuracy value from confusionMatrix in R?

In R caret library, if I got a confusion matrix like this below, if there a way to retrieve the overall accuracy 0.992? I can't get this single value out, since I need to store this value and use it for later processing. Is this possible at all?

Prediction A B C D E
A 1114 2 0 0 0
B 9 745 5 0 0
C 0 6 674 4 0
D 0 0 3 640 0
E 0 0 2 1 718


Overall Statistics

Accuracy : 0.992
95% CI : (0.989, 0.994)
No Information Rate : 0.286
P-Value [Acc > NIR] : <2e-16

Kappa : 0.99


Mcnemar's Test P-Value : NA

Statistics by Class:

Class: A Class: B Class: C Class: D Class: E
Sensitivity 0.992 0.989 0.985 0.992 1.000
Specificity 0.999 0.996 0.997 0.999 0.999
Pos Pred Value 0.998 0.982 0.985 0.995 0.996
Neg Pred Value 0.997 0.997 0.997 0.998 1.000
Prevalence 0.286 0.192 0.174 0.164 0.183
Detection Rate 0.284 0.190 0.172 0.163 0.183
Detection Prevalence 0.284 0.193 0.174 0.164 0.184
Balanced Accuracy 0.996 0.992 0.991 0.996 1.000

Answer

Given a confusion matrix cm, the overall accuracy is obtained by overall.accuracy <- cm$overall['Accuracy']

It's the first time I see the caret package, so how did I know this?

Since you didn't provide an example, I searched for an example code for caret confusion matrices. Here it is (I only added assignment in the last statement):

###################
## 3 class example

confusionMatrix(iris$Species, sample(iris$Species))

newPrior <- c(.05, .8, .15)
names(newPrior) <- levels(iris$Species)

cm <- confusionMatrix(iris$Species, sample(iris$Species))

Now, let's take a look what's in the confusion matrix:

> str(cm)
List of 5
 $ positive: NULL
 $ table   : 'table' int [1:3, 1:3] 13 18 19 20 13 17 17 19 14
  ..- attr(*, "dimnames")=List of 2
  .. ..$ Prediction: chr [1:3] "setosa" "versicolor" "virginica"
  .. ..$ Reference : chr [1:3] "setosa" "versicolor" "virginica"
 $ overall : Named num [1:7] 0.267 -0.1 0.198 0.345 0.333 ...
  ..- attr(*, "names")= chr [1:7] "Accuracy" "Kappa" "AccuracyLower" "AccuracyUpper" ...
 $ byClass : num [1:3, 1:8] 0.26 0.26 0.28 0.63 0.63 0.64 0.26 0.26 0.28 0.63 ...
  ..- attr(*, "dimnames")=List of 2
  .. ..$ : chr [1:3] "Class: setosa" "Class: versicolor" "Class: virginica"
  .. ..$ : chr [1:8] "Sensitivity" "Specificity" "Pos Pred Value" "Neg Pred Value" ...
 $ dots    : list()
 - attr(*, "class")= chr "confusionMatrix"

As you may see, the cm object is a list. We see various "byClass" and "overall" statistics. The overall part is obtained by:

overall <- cm$overall

Which gives us a vector of numbers with string indices:

> overall
      Accuracy          Kappa  AccuracyLower  AccuracyUpper   AccuracyNull AccuracyPValue  McnemarPValue 
     0.2666667     -0.1000000      0.1978421      0.3449492      0.3333333      0.9674672      0.9547790 

Now, extracting the relevant value is as simple as:

> overall.accuracy <- overall['Accuracy'] 

Summary: str is your friend. Another useful function is attributes -- it returns all the attributes of a given object.

Comments