Confusion Matrix give information on misclassification. How patterns have been classified in their actual classes against other classes for each class in a matrix form.
The author does not preallocate. This would be a faster engine (and there's probably a faster bsxfun solution)
L = length(c);
conf2 = zeros(L);
for ii = 1:L
Tii = T==ii;
S = sum(Tii);
for jj = 1:L
conf2(ii,jj) = sum ((Y==jj).*Tii)/S*100;
The help is wrong here:
"The vector RATE has two entries: the percentage of correct classifications and the total number of correct classifications."
RATE is actually returned as a scalar.