What should a classifier system learn?
Tim Kovacs · 2002
We consider the issue of how a classifier system should learn to represent a Boolean function. We identify four properties which may be desirable of a representation; that it be complete, accurate, minimal and nonoverlapping, and distinguish variations on two of these properties for the XCS system. We question whether the bias against overlapping rules evident in some systems is appropriate, and find that XCS's bias against overlapping rules is very strong.