The most significant modification to NEFCLASS was to introduce a matrix containing the misclassification costs. This is basically a n by (n+1) matrix, where the represent the costs caused if the system classifies a pattern of class i as class j, or as ambiguous (j=n+1). The diagonal elements are usually equal to 0, all other elements are set to 1 by default. If the domain bears asymmetries of the classes this matrix allows rather fine and intuitive adaption of the errors.
The next sections describe the modifications, that had to be performed to take this misclassification cost matrix into account.