Heuristic-based Genetic Operation in Classifier Systems

This thesis focuses on improving the Accuracy-based Learning Classifier System (XCS), a Machine Learning technique that attempts to build general and accurate rules. Adapted from the induction concept, a new approach named Rule Combining (RC) draws conclusions from the experience. It learns more efficient in terms of speed and space requirement compared to the original XCS that employs Darwinian genetic operation. Furthermore, RC allows an additional capability of performing feature selection.