Towards generalization by identification-based XCS in multi-steps problem
Masaya Nakata, Fumiaki Sato, Keiki Takadama · 2011
This paper extends an accuracy-based Learning Classifier System (XCS) to promote a generalization of classifiers by selecting effective ones and deleting ineffective ones, and calls it Identification-based XCS (IXCS). Through the intensive simulations of the Maze problem (Maze6), the following implications have been revealed : (1) IXCS can derive good solutions with a fewer number of classifiers in comparison with XCSG as one of the major conventional XCS; and (2) IXCS can not only generalize the classifiers faster but also generate the classifiers that are robust to the noisy environment.