Based on the analysis of features of the grid-based clustering method-clustering in quest (CLIQUE) and density-based clustering method-density-based spatial clustering of applications with noise (DBSCAN), a new cl...Based on the analysis of features of the grid-based clustering method-clustering in quest (CLIQUE) and density-based clustering method-density-based spatial clustering of applications with noise (DBSCAN), a new clustering algorithm named cooperative clustering based on grid and density (CLGRID) is presented. The new algorithm adopts an equivalent rule of regional inquiry and density unit identification. The central region of one class is calculated by the grid-based method and the margin region by a density-based method. By clustering in two phases and using only a small number of seed objects in representative units to expand the cluster, the frequency of region query can be decreased, and consequently the cost of time is reduced. The new algorithm retains positive features of both grid-based and density-based methods and avoids the difficulty of parameter searching. It can discover clusters of arbitrary shape with high efficiency and is not sensitive to noise. The application of CLGRID on test data sets demonstrates its validity and higher efficiency, which contrast with tradi- tional DBSCAN with R tree.展开更多
在优化空间聚类算法的研究中,传统的K-means空间算法存在两个缺陷,其一是对空间对象的属性描述不全面,其二是对初始种子集选取敏感,容易陷入局部最优值,聚类结果不稳定。为了优化算法,引入适合空间对象的空间属性距离和基于最大维密度...在优化空间聚类算法的研究中,传统的K-means空间算法存在两个缺陷,其一是对空间对象的属性描述不全面,其二是对初始种子集选取敏感,容易陷入局部最优值,聚类结果不稳定。为了优化算法,引入适合空间对象的空间属性距离和基于最大维密度选择方案(Max-Dimension of Density Based Seeking,MDDBS)来改进K-means算法,提出利用最大维密度的全局优化空间聚类算法(Max-Dimension of Density Based Clustering,MDDBC),可从密度大的区域选取初始种子,同时又尽量将种子分散在数据空间。实验结果表明,改进方法可以很好消除聚类结果的波动性,同时更加客观地呈现空间对象的分布规律。展开更多
基金This project is supported by National Natural Science Foundation of China(No.50575153).
文摘Based on the analysis of features of the grid-based clustering method-clustering in quest (CLIQUE) and density-based clustering method-density-based spatial clustering of applications with noise (DBSCAN), a new clustering algorithm named cooperative clustering based on grid and density (CLGRID) is presented. The new algorithm adopts an equivalent rule of regional inquiry and density unit identification. The central region of one class is calculated by the grid-based method and the margin region by a density-based method. By clustering in two phases and using only a small number of seed objects in representative units to expand the cluster, the frequency of region query can be decreased, and consequently the cost of time is reduced. The new algorithm retains positive features of both grid-based and density-based methods and avoids the difficulty of parameter searching. It can discover clusters of arbitrary shape with high efficiency and is not sensitive to noise. The application of CLGRID on test data sets demonstrates its validity and higher efficiency, which contrast with tradi- tional DBSCAN with R tree.
文摘在优化空间聚类算法的研究中,传统的K-means空间算法存在两个缺陷,其一是对空间对象的属性描述不全面,其二是对初始种子集选取敏感,容易陷入局部最优值,聚类结果不稳定。为了优化算法,引入适合空间对象的空间属性距离和基于最大维密度选择方案(Max-Dimension of Density Based Seeking,MDDBS)来改进K-means算法,提出利用最大维密度的全局优化空间聚类算法(Max-Dimension of Density Based Clustering,MDDBC),可从密度大的区域选取初始种子,同时又尽量将种子分散在数据空间。实验结果表明,改进方法可以很好消除聚类结果的波动性,同时更加客观地呈现空间对象的分布规律。