I mean the accuracy of the classifier's prediction of payoff.

In XCS, each classifier keeps a record of its prediction of payoff, and the average error in that prediction. The classifier's accuracy is then defined as a particular inverse function of the error, as discussed in Section 3.4.

What do you mean by a classifier's "accuracy"?