Dear Colleagues, I am using a Bayesian network as a classifier. The class variable are four-folds. With a binary class variable I can use the ROC curve to plot true-positive rate versus false-positive rate. This acronym stands for "receiver operating characteristic". With two-folds variabels, in the medical field, the area under the ROC curve is sometimes understood as a way to summarize the diagnostic discrimination. This is also called the probability of concordance between predicted and observed disease states.
It is easy to extend the calculation method to deal with n-folds variables but is it reasonable? What would be the semantic of the area under the extended ROC curve? I'd be grateful for any pointers to this question. regards, Marcelo. ----------- Prof. Dr. Marcelo Ladeira Computer Science Department University of Brasilia, Brasilia, Brazil Caixa Postal 4466 - CEP 70.919-970 Phone +55 (61)307-2702 ext. 213 Fax +55 (61)273-3589, 273-21-31 email: [EMAIL PROTECTED] URL: http://www.cic.unb.br/~mladeira
