I am currently doing a project in college. Evaluating the KDD 99 set using Neural Network. I am using nnet in R library. However, the KDD 99 have 23 types of attack, my net is learning to classify it. How can I evaluate the accuracy of it? I mean a confusion matrix is meant for 2 class problem, if not, max 3 class. Is there a way I can evaluate my nnet model? Thanks.
Asked
Active
Viewed 769 times
0
-
1Cross-validated may be more appropriate for this question – Whitebeard Aug 27 '15 at 13:19
-
Confusion matrix by definition just needs the actual classes and the predicted classes. There's no maximum threshold. Of course, it's more common to see a confusion matrix for 2 and 3 classes, because it's easier to interpret. You can use it for multiple classes and, obviously, the accuracy is better if the observations are in (or around) the diagonal elements. – AntoniosK Aug 27 '15 at 13:28