Multiclass error area under the ROC
E = TESTAUC(A*W)
The area under the ROC is computed for the datset A w.r.t. the classifer W. The estimator is based on a rank analysis of the classifier outcomes. Ties are broken by a two-way sorting and averaging.
The multiclass situation is solved by averaging over all outcomes of the one-against-rest ROCs.
Note that E is an error and not a performance measure like the AUC often used in literature.