Clustering with XCS on Complex Structure Dataset

Learning Classifier System (LCS) is an effective tool to solve classification problems. Clustering with XCS (accuracy-based LCS) is a novel approach proposed recently. In this paper, we revise the framework of XCS, and present a complete framework of clustering with XCS. XCS consists of two major modules: reinforcement learning and genetic algorithm. After the learning process, the learnt rules are always redundant and the large ruleset is incomprehensive. We adopt the revised compact rule algorithm to compress the ruleset, and propose a new rule merging algorithm to merge rules for generating genuine clustering results without knowing of the number of clusters. The experiment results on several complex structure datasets show that out approach performs well on challenging synthetic datasets.

• Type: Conference
• Year: 2008
• Where: AUSAI
• Authors: Liangdong Shi, Yang Gao, Lei Wu, Lin Shang
• Keywords: Artificial Intelligence, AUSAI 2008, Compact Rule Algorithm, Learning Classifier System, Rule Merging Algorithm