论文标题:聚类分析在暂住人口分析中的应用研究
The Research and Application on the Population Clustering Algorithm for Data Mining
论文作者
论文导师 田忠和,论文学位 硕士,论文专业 控制科学与工程系
论文单位 华中科技大学,点击次数 86,论文页数 76页File Size817K
2006-05-01论文网 http://www.lw23.com/lunwen_681339737/
data mining;; clustering analysis;;“shack up in”population;; k-medoids algorithm;; difference matrix;; CLIQUE algorithm
人口资源是最具战略性的资源,作为世界上人口最多的国家,加强人口管理现代化,对于我们国家各项事业的发展至关重要,其中暂住人口管理就关系着治安管理。在数据挖掘领域中,聚类分析是一项重要的研究课题,聚类的目标是在没有任何先验知识的前提下,根据数据的相似性将数据聚合成不同的簇,使得相同簇中的元素尽可能相似,不同簇中的元素差别尽可能大。把聚类分析应用到暂住人口的挖掘中,可以发现不同特征的暂住人群,对暂住人口的调整和控制有很大的帮助。 目前,人们提出了很多种聚类算法。其中,典型代表是基于距离的聚类算法和基于密度网格的聚类算法。本文主要研究以k-中心点算法为代表的基于距离的聚类算法和以CLIQUE算法为代表的基于密度和网格的聚类算法。在k-中心点算法中,探讨用中心点差异矩阵来提高聚类速度,避免了传统算法中迭代次数随数据记录以平方增长,不适应大数据集的弊端。在CLIQUE算法中,改进密度阀值和区间个数单一的做法,提高聚类精确度:细化属性分区,不同的属性区间个数不同;随维数的增加,密度阀值相应减小,以获得更大的聚类覆盖面。 为了测试改进的聚类算法的性能,本文设计并实现了一个关于暂住人口的聚类实验,程序可以方便地实现数据源连接、聚类分析以及二维数据的可视化。实验验证了k-中心点改进算法可以快速聚类,但是容易陷入局部最优解。而优化后的CLIQUE实现了更大的覆盖。最后用改进的k-中心点算法思想和CLIQUE的基于属性区间统计思想解决了一个实际研究课题。
As population resource is the most important one for a country, reinforcing population management with information tools is very important for other projects in our country with the largest population in the world.In more detail, The“shack up in”population is related with the public security management. Clustering analysis is an important research problem in the domain of data mining. The goal of clustering is to partition data set into such clusters that intra-cluster data are similar and inter-cluster data are dissimilar without any prior knowledge, which is very different from data classification. Clustering analysis on the“shack up in”population and discovering different“shack up in”groups are helpful to the modification and establishment of the public security management. Many clustering algorithms are presented which include distance-based clustering algorithms, density-based clustering algorithms and grid-based clustering algorithms. This paper mainly researches on distance-based clustering algorithms whose representative is k-medoids algorithm and density-grid-based clustering algorithms whose representative is CLIQUE algorithm. In k-medoids algorithm the times of iterative loop is increasing square as the number of data increasing. Therefore The article discusses a difference matrix of the medoids to improve the speed of clustering. In CLIQUE algorithm the density valve has only one value in despite of the dimension increasing,ξ(the partition number at every dimension) only one too. The optimization methods in the article can make the data overlap maximized: every dimension has its ownξand the density valve decreasing as the dimension increasing. In order to test the performance of clustering algorithms, we design and realize a clustering experimental program, which carries out data connection, clustering and two-dimensional data visualization. Experimental results show that the improved k-medoids algorithm is fast but easy to get local optimization. And the optimized CLIQUE algorithm gets larger scale. At last, a practical problem is solved by using the idea of ameliorating k-medoids algorithm.
|
