Table III presents the average confusion matrix of recognition of all classes of coffee blends prepared for the testing data
in all 30 runs of the algorithm. The diagonal elements of this
matrix represent the percentage of well-recognized classes. The
nonzero value of the ijth off-diagonal element indicates what
percentage of the ith class was recognized as the jth one. Its
observed relative mean value over all classes was equal 0.21%.
In half of the runs, there were no misclassifications at all.
Among 1155 cases used in testing, only few testing data have
been misclassified in some runs. However, even in these cases,
the misclassifications were between two neighboring classes