Thankyou Mr.Waldemar. Gnuplot is a poting tool, but my question is how to get TPR and FPR values from confusion matrices to draw a ROC curve without prediction probabilities.

To my point of view ROC helps mostly for binary classification analysis.

However, there are various ways to plot it for multi-label case, too. For python, you have a nice example in scikit-learn, which uses averaging trick. See example with code here: http://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html

In my case, I have a multi class matrix confusion and I plot a curve for each class in the same plot. You would check this paper with an approach by pairwise analysis.

http://homepage.tudelft.nl/a9p19/papers/prl_07_roc.pdf

Thank you @Arseniy Gorin and @John Jairo Villarejo Mayor

Dear Rajesh N V P S Kandala ,

Great discussion point. Following the updates!

Your answer is well-documented in the scikit-learn lib.

Basicly, the solution for multiclass is to binarize the dependent variable, then the problem naturally disappears.

Hope it helps:

https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html

Check this example

scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html

Thank you all for your thoughtful suggestions.