欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
提出了一种有效选择初始聚类中心的算法CNICC．该算法参考了网格聚类算法的思路．划分数据空间为相应维度上的网格单元，然后根据实例的分布情况确定初始聚类中心．从二阶差分的概念出发，CNICC定义了网格单元的一阶邻居和二阶邻居，算法根据每个网格单元的一阶和二阶邻居的局部密度变化寻找包含聚类中心的网格单元．在人工数据集上进行的实验表明，与现有初始化聚类中心的方法相比，CNICC能够有效减少K-means算法的迭代次数，提升聚类精度．同时，随着数据集实例数、维度和网格单元数量的增加，算法的时间复杂度呈线性增加．
英文摘要：
In this paper, a new method is presented for effectively selecting initial cluster centers, CNICC. This method adopts the general idea of grid-based clustering algorithm. It firstly partitions the data space into some grid cells of corresponding dimensions, and then selects the initial cluster centers according to the distribution of instances. Considering the concept of second-order difference, CNICC gives the definitions of first-order neighbor and second-order neighbor. According to the local density tendency of both first-order neighbor and second-order neighbor of each gird cell, CNICC searches for the grid cells with cluster centers. Experiments on synthetic datasets show that compared with current approaches, CNICC is able to effectively decrease the number of K-means iterations and improve clustering precision. Moreover, the running time of CNICC is linear with respect to the number of instances, the number of grid cells and the dimensions.
同期刊论文项目
同项目期刊论文
期刊信息
位置：