6

When I understand it correctly, a ROC curve composes of a lot of ratios of TPR and FPR. Each such ratio is gained, e.g., by scanning/investigating a parameter where the ratio is the result of the used classifier. In other words: I'm interested in how a classifier separates data points according to a frequency interval of 0-100 Hz. Hence I scan this frequency, record all the 101 ratios, put them into the plot and the ROC curve represents these datapoints, finally.

img

Wouldn't it be much more interesting when the frequency would be used as a third axis? Or is it implicit in the axis? But anyway, I think it would be easier to look at with a third axis?

Dave
  • 62,186
Ben
  • 3,443
  • Your threshold strictly determines FPR and TPR so it wouldn't add much new information and it'd make the chart more harder to interpret – David Jun 16 '21 at 14:31
  • 1
    My question (5 days ago) was closed, because someone else asked the same (2 days ago)? :D – Ben Jun 22 '21 at 06:38

1 Answers1

5

The ROC curve is two-dimensional because it is defined as the plot of TPR and FPR. Yes, you might be interested in the threshold used to generate the TPR and FPR, and you might want to plot that on a third axis, but then it is not a ROC curve but something else (arguably more informative, arguably harder to draw, arguably so hard to draw that the additional information is not worth the trouble).

Dave
  • 62,186
  • 1
    Especially considering it's most likely going to be displayed on a 2-D screen or printed on 2-D paper... – Calimo Jun 16 '21 at 18:14