َIs it correct to speak of receiver operating curve (ROC) for k-nearest neighbor algorithm?

As far as I know, the ROC is plotted by changing the decision threshold, calculating FPR and TPR for each value of the decision threshold, and plotting the (FPR, TPR) points. I understand how that process is done in, for example, a Bayesian decision-making algorithm since it is clear that what is the "decision threshold" the Bayesian decision-making theory (the decision threshold locates on the point where the two discriminant functions are equal). Regarding that, I cannot understand what is the decision threshold in the KNN algorithm. without any decision threshold, how it is possible to draw a ROC? I will appreciate it if somebody could answer me.

from Recent Questions - Stack Overflow https://ift.tt/3nvVmu4
https://ift.tt/eA8V8J