Effective XCS Search: Building Block Processing

Abstract

The facetwise approach to GA theory stresses effective mixing and decision making among BBs. The last chapter showed that in XCS, BBs are subsets of specified attributes that increase accuracy. The reproductive opportunity bound additionally ensures that BBs are able to grow in the population making time for the identification and reproduction of schema representatives. Until now, we assumed that mutation is sufficient to generate better classifiers as investigated in the time bound. However, the GA literature suggests that effective crossover operators are mandatory to solve boundedly difficulty optimization problems in which small, lower-level BBs may mislead the population to a local optimum.