Earlier and some current classifier systems use strength both for the payoff prediction and as the classifier's fitness. This includes systems described by Holland, at least since the bucket brigade was invented. However, Holland at first had fitness depend not only on payoff, but also on the consistency of the payoff. And Booker has recently used a composite fitness measure.

Since the appearance of the present paper, "Classifier fitness based on accuracy", several classifier systems have followed the paper's scheme in which strength is dropped, and payoff prediction and classifier fitness are treated as separate quantities.

When you say "traditionally", do you mean most classifier systems are like that? Which ones do not use strength, or do not use it that way?