Learning overlapping natured and niche imbalance boolean problems using XCS classifier systems

XCS is an accuracy-based learning classifier system, which has been successfully applied to learn various classification and function approximation problems. Recently, it has been reported that XCS cannot learn overlapping natured and niche imbalance problems using the typical experimental setup. Previously we have developed an XCS with code-fragment action, named XCSCFA, which has the unusual property that during training the action value in a classifier rule can vary, even for the same problem instance, at different times. In the work presented here, the XCSCFA approach is applied to four different complex Boolean problem domains including the overlapping natured and niche imbalance domains. The XCSCFA system successfully learnt all the experimented problems. The major contribution of this work is overcoming the identified problem in the widespread XCS technique, i.e. it is no longer impossible to learn overlapping natured and niche imbalance problems.