摘要
分析了目前基于目标函数聚类算法的不足,面对形状复杂且非重叠的样本聚类问题,定义了最邻近距离和生长树的概念。随机选取生长树初始种子点,以最邻近距离作为生长树生长的方向和样本划分依据,以最终生长树大小为聚类目标函数,引入遗传算法,提出基于生长树的遗传聚类算法,并通过实例进行了算法测试和比较。算法测试表明:基于生长树的遗传聚类算法对于形状复杂且非重叠样本的聚类是完全可行和有效的。
The shortcomings about these days clustering algorithm based on aim function are analyzed. In order to dealing with the clustering of complex shape and no-overlap samples, the concept of the best-close distance and propagating tree are defined. The genetic-clustering algorithm based on the propagating tree is put forward, selecting randomly the initialization seed points of propagating tree, making the propagating direction of propagating tree and partitioning samples according to the bestclose distance, calculating cluster aim function on propagating tree value, importing genetic algorithm. The algorithm is validated and compared with others by examples. Algorithm testing show that it is completely feasible and availability for the genetic-clustering algorithm based on the propagating tree to deal with the clustering of complex shape and no-overlap samples.
出处
《计算机应用研究》
CSCD
北大核心
2006年第7期62-64,共3页
Application Research of Computers
基金
国家自然科学基金资助项目(60202004)
关键词
聚类算法
数据挖掘
生长树
遗传算法
Clustering Algorithm
Data Mining
Propagating Tree
Genetic Algorithm