Abstract
We consider the issues of how a classifier system should learn to represent a Boolean function, and how we should measure its progress in doing so. We identify four properties which may be desirable of a representation; that it be complete, accurate, minimal and non-overlapping, and distinguish variations on two of these properties for the XCS system. We distinguish two categories of learning metric, introduce new metrics and evaluate them. We demonstrate the superiority of population state metrics over performance metrics in two situations, and in the process find evidence of XCS's strong bias against overlapping rules.
| Translated title of the contribution | What should a classifier system learn? |
|---|---|
| Original language | English |
| Title of host publication | Unknown |
| Editors | Jong-Hwan Kim |
| Publisher | Institute of Electrical and Electronics Engineers (IEEE) |
| Pages | 775 - 782 |
| Number of pages | 7 |
| Publication status | Published - May 2001 |