基于复杂学习分类系统的密度聚类方法

发布时间：2018-05-27 00:28

本文选题：学习分类系统 + 进化计算　；参考：《计算机应用》2017年11期

【摘要】：提出一种基于复杂学习分类系统(XCS)的密度聚类方法,可以用于对任意形状且带有噪声的二维数据进行聚类分析。此方法称为DXCSc,主要包括以下三个过程:1)基于学习分类系统,对输入数据生成规则种群,并对规则进行适当压缩;2)将已经生成的规则视为二维数据点,进而基于密度聚类思想对二维数据点进行聚类;3)对密度聚类后的规则种群进行适当聚合,生成最终的规则种群。在第一个过程中,采用学习分类系统框架生成规则种群并进行适当约减。第二个过程认为种群的各规则簇中心比它们的邻居规则具有更高的密度,并且与密度更高的规则间距离更大。在第三个过程中,采用图分割方法对相关重叠簇进行适当聚合。在实验中,将所提方法与K-means、近邻传播聚类算法(AP)、Voting-XCSc等算法进行了比较,实验结果表明,所提方法在精度方面优于对比算法。
[Abstract]:A density clustering method based on complex learning classification system (XCS) is proposed, which can be used for clustering analysis of two dimensional data with arbitrary shape and noise. This method, called DXCSc, mainly consists of the following three processes: 1) based on the learning classification system, the rule population is generated for the input data, and the rules are appropriately compressed) the rules that have been generated are regarded as two-dimensional data points. Then, based on the idea of density clustering, the two-dimensional data points are clustered by 3) the regular population after density clustering is properly aggregated, and the final regular population is generated. In the first process, the learning classification system framework is used to generate the rule population and reduce the rule population. The second process considers that the regular cluster centers of the population have higher density than their neighbor rules, and the distance between the regular cluster centers and the higher density rules is larger. In the third process, the graph segmentation method is used to polymerize the overlapped clusters. In the experiment, the proposed method is compared with K-means, nearest neighbor propagation clustering algorithm and Voting-XCSc algorithm. The experimental results show that the proposed method is superior to the contrast algorithm in accuracy.
【作者单位】：计算机软件新技术国家重点实验室(南京大学);江苏省审计厅;
【基金】：江苏省重点研发计划(产业前瞻与共性关键技术)项目(BE2015213)~~
【分类号】：TP181

【相似文献】