Hi,
Is there anyone that can help me soon. I have to finish this part of my Master project urgently!
I used the three prediction models (DT, SVM, and naive Bayes). I faced two issues:
1- How can I plot the ROC curve for the SVM model?
2- How can I plot all three model results only on one ROC?
Thank you and regards,
Vida
I have attached a sample workflow for 3 scores and a ROC curve plot in two different ways.
Maybe you could adapt that example.
kn_example_roc3.knar (405.2 KB)
Thank you so much. I will try it right now.
It works 
thanks!
Do you know why It is not possible to show the SVM result on ROC? Is there any issue with binary classification?
Great if it does work in the first place.
Could you give an example or a screenshot how it looks. Is the SVM result a Score between 0 and 1? Or is ist just a prediction of 0 or 1?
Yes, it shows prediction like 0 or 1. I attached the table and ROC results.
The other predictive models like DT and NB are shown binary like 0 or 1 but the final result on ROC is like the percentage, not 0 or 1. Do you know why?
The ROC actually only works for binary classification.
In the dialog you have to specify the column containing the ground truth (i.e. the actual classes of the rows) and the column containing the predicted probability for this class.
From your screenshot it seems like you selected both the probability for the positive and for the negative class.
Your model also seems to be a bit too good because it has a AUC of 1 which is in most cases too good to be true unless you predict the training data.
Cheers,
nemad
For a binary classification task, an AUC of 0.5 corresponds to a model that randomly draws one of the two classes which obviously is not desirable. If a model has an AUC below 0.5, as it is the case for your SVM, this usually implies some issue because inverting the model’s predictions would actually result in an AUC greater than 0.5 (also see this CrossValidated question).
Perhaps you accidentally use the probability for the negative class in the ROC?
In case of more than two classes, it is more difficult to draw a conclusion because the ROC still considers a binary classification problem where all classes that are not the positive class are considered to be in the negative class.
Thank you so much for your kind support.
