0

I have read this question but it doesn't have any well-explained answer for this case: I understand the ROC curve in overall, but I'm looking for a step-to-step explanation in order to understand how the curve changes when one true-positive becomes false-positive (and so on). For example, how can I draw a ROC curve with this set?. Does the score means the "certainty" for the value given? (for example: for the id == A the is a 0.03 of certainty that the value is active/decoy == a?)

id  score   active/decoy 
A   0.03    a
B   0.08    a
C   0.10    d
C   0.11    a
D   0.22    d
E   0.32    a
F   0.35    a
G   0.43    a
H   0.57    a
I   0.61    d
J   0.68    a
kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Chris
  • 171
  • 1
  • 8
  • 1
    Did you read this question? https://stats.stackexchange.com/questions/145566/how-to-calculate-area-under-the-curve-auc-or-the-c-statistic-by-hand Can you be more specific about which part you do not understand? – Sycorax Feb 20 '20 at 21:09
  • 4
    Start with a cutoff of 0.001. If the score is above the cutoff, mark is "d". Otherwise, mark it "a". Determine the sensitivity and 1-specificity. Do it again for a cutoff of 0.002. Do it again for a cutoff... – Dave Feb 20 '20 at 21:09

1 Answers1

0

Here is a general algorithm I have used in the past to draw ROC

  1. Sort the data by score from lowest to highest
  2. Choose lowest score as the cut-point
  3. Calculate sensitivity and specificity using the cut-point from step 2. Save sensitivity and 1-specificity
  4. Repeat steps 2 and 3 for each unique score from lowest to highest
  5. Create a step plot where the x-axis is 1-specificity and the y-axis is sensitivity

I would recommend re-coding active/decoy as 0/1 for the algorithm

pzivich
  • 1,430
  • 1
  • 5
  • 15