摘要
多目标聚类过程中会产生一些明显不合理的解，影响最终划分结果以及聚类类数的判断。为此，提出一种基于局部集成和克隆选择的多目标聚类算法。在聚类过程中周期性的将聚类解集划分为若干邻域，对每个邻域进行局部集成操作，剔除各个类数下的不舍理划分；利用克隆选择算法的思想构建3种变异算子，推动种群的进化，分别具有增大或减小当前解的聚类类数、调整当前解样本划分情况的功能。3组人工数据集以及3组UCI数据集的实验结果表明，该算法能够得到优于对比算法的聚类结果，准确判断出合理的聚类类数，判断类数的准确率可提高0％～46．67％。
A few obviously infeasible solutions usually exist in multi-objective clustering, which is harmful for selecting the best solution and identifying the number of clusters. Hence, a multi-objective clustering algorithm based on local ensemble and clone selection was proposed. The solution set was periodically divided to several neighboring subsets and then local ensemble was made on each of them. In this way, these infeasible solutions with different numbers of clusters were removed from current solu tion set. Besides, three mutation operators were designed for clone selection to improve the evolution process. Results of experi- ments on three synthetic and three UCI datasets show that the proposed algorithm obtains better clustering results than other al- gorithms and determines the right number of clusters most of the time, and the accuracy of determination is promoted by 0%-46.67%.
出处
《计算机工程与设计》
北大核心
2015年第8期2234-2238,共5页
Computer Engineering and Design
基金
国家自然科学基金青年基金项目(50904067)
关键词
多目标聚类
局部集成
克隆选择
聚类类数
种群进化
multi-objective clustering
local ensemble
clone selection
number of clusters
population evolution
