I'm trying to figure out how to present results (e.g. precision, recall, accuracy, ROC curve) for a class project.
I have a multi-class classification problem(specific classification algorithm not really important) of classes A, B, C and D.
The classification algorithm was trained on A, B and C.
During the test phase, classification algorithm would determine unit test data to be class A if likelihood_of_test_data >= likelihood_threshold_of_A which is obtained in training phase.
Similarly classes B and C are also recognized in test phase.
If likelihood_of_test_data < likelihood threshold of A, B and C, it would be recognized as belonging to class D.
How can I draw the ROC curve in this case? Do I have to calculate precision and recall for each of the classes of A, B, C and D separately? Or I calculate just the accuracy?
