关键词: 学习分类系统, 进化计算, 强化学习, 密度聚类, 规则合并

Abstract: In this paper, we propose a density clustering method based on complex real Learning Classifier System (XCS), which can be used to cluster the two-dimensional data sets with different shapes and noises. This method is called Density of XCS clustering (DXCSc), which mainly includes the following three processes: (1) Generate and compact the rule sets of the population following the learning classification system. (2) Based on the idea of density clustering, the rules sets that have been generated are treated as two-dimensional data points, and then the rule population sets are clustered. (3) The rule population sets after density clustering are properly merged to generate the final rule sets of population. In the first process, the Learning Classifier System framework is used to generate and compact the regular population. In the second process, we consider that the rule cluster centers are characterized by a higher density than their neighbors and by a relatively large distance from points with higher densities. In the third process, the relevant clusters are properly merged using the graph segmentation method. In our experiments, we compared the proposed DXCSc with K-means AP and Voting-XCSc on a number of challenging data sets. The results show that the proposed approach outperforms K-means and Voting-XCSc in the precision rate.

Key words: Learning classifier systems, evolutionary computing, reinforcement learning, density clustering, rule merging