How to plot ROC curve?
8 visualizzazioni (ultimi 30 giorni)
Mostra commenti meno recenti
I have dataset which I classified using 10 different thresholds. Then I evaluated true and false positive rate (TPR, FPR) to generate ROC curve. However, the curve looks strange. Did I evaluated the curve correctly? Below is the code which I used to generate ROC curve.
TPR=[0.214091009346534 0.231387608987612 0.265932891531049 ...
0.324782536928746 0.407704239947213 0.497932979272465 ...
0.566189022386499 0.587833185570207 0.546182718263242 ...
0.434923996561788];
FPR=[0.006017495627892 0.008669605012233 0.013377312018797 ...
0.022621821298088 0.039994426565193 0.069264094928662 ...
0.108694153334795 0.148784394110204 0.178634096117665 ...
0.194756822274831];
plot(FPR,TPR);
2 Commenti
Cretu Calin
il 7 Apr 2017
I think that the last two values are wrong. You cannot have any point in the right side of the diagonal [(0,0),(1,1)].
Image Analyst
il 7 Apr 2017
Cretu, please explain why believe the last two points are to the right of the 0-to-1 diagonal:
Risposta accettata
Thorsten
il 25 Nov 2015
I agree that the curves look strange. If you decrease the threshold, you cannot get a smaller true positive rate. The rate can only stay the same or increase. So your two points at the end of the curve are wrong. Also, you should vary your threshold through the full range, from max to 0, such that your curve starts from (0,0) and ends at (1,1).
9 Commenti
Thorsten
il 25 Nov 2015
Modificato: Thorsten
il 25 Nov 2015
You have to restrict all computations to the valid indices:
valid_ind = R > -3.40e+38;
P = nnz(GT(valid_ind)); % number of positive responses in ground truth
N = nnz(1-GT(valid_ind));
and in the loop
tp(i) = nnz(Rt(valid_ind) & GT(valid_ind));
fp(i) = nnz(Rt(valid_ind) & ~GT(valid_ind));
The best way to do it is to write a function
function [TPR, FPR] = roc(GT, R, thresholds)
and call this function with
GT(valid_ind), R(valid_ind)
in case you have to exclude some pixels from the analysis.
Natsu dragon
il 3 Feb 2018
Modificato: Natsu dragon
il 3 Feb 2018
hello, i have used the same code with your attached data and i got results, but when i used it with my own results i got nothing, it didn't plots. can you help me to understand why?
Più risposte (0)
Vedere anche
Categorie
Scopri di più su Detection in Help Center e File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!