8
$\begingroup$

Most of the confusion matrices I've seen contain the number of instances in each cell. Isn't a confusion matrix with the percentage of instances in each cell easier to read? Is this approach wrong or does it go against some unwritten rule with regards to confusion matrices?

Such an confusion matrix will look like this, where each of the 10 class labels makes up 10 percent of the dataset and the total is 100 percent. 9.06 percent of the dataset belonged to class 1 and was assigned to class 1. Therefore 90.60 percent of class 1 instances are classified correctly.

example

Improve this question
$\endgroup$
2
  • 4
    $\begingroup$ A percentage without an absolute magnitude or a single magnitude without an overall measure of scale both lack part of the picture, so in that respect neither is an ideal solution. I don't see anything wrong with percentages if they convey better what it is you wish to say or interpret. $\endgroup$
    – image_doctor
    Commented Apr 8, 2013 at 13:14
  • $\begingroup$ Note that with instance counts, you can easily compute percentages. With percentages, you also need the total number of instances and potentially introduce some rounding errors. $\endgroup$
    – Gala
    Commented Jun 8, 2013 at 6:29

1 Answer 1

Reset to default
3
$\begingroup$

A confusion matrix in percents would be appropriate if the distribution between your classes is flat (either naturally, or intentionally sampled that way). If this is not the case, such a confusion matrix can lead to major confusion.

It is useful to have both: number of instances overall (to see the skews) and percents for data sampled from a flat distribution.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.