论文增量聚类算法的设计与实现.doc

 增量聚类算法的设计与实现 摘 要 随着网络技术和数据库技术的发展，各个领域的信息量急剧增长，需要处理的数据规模越来越大；同时，对数据集实时处理的需求不断加大。这就促使了海量动态数据的出现。目前大多数聚类算法处理海量动态数据不仅需要消耗大量的时间和空间资源，并且挖掘效果不理想，其有效性和可扩展性受到了限制。在这种情况下，增量聚类算法应运而生，有效的解决了海量动态数据的聚类问题。 为了实现动态、增量聚类，首先，对基于密度和自适应密度可达聚类算法（Clustering Algorithm Based on Density and Density reachable, CADD）进行了改进：第一，为密度可达包成员设置标志值，极大的提高了聚类效率；第二，改进半径和密度的计算方式，避免了重复的计算；第三，编程实现聚类结果的可视化，可以直观的评价聚类效果。实验结果表明，改进后使得算法的复杂度明显减小。 其次，在CADD算法改进的基础上重点研究了两方面内容：（1）根据CADD算法的特点，提出了基于密度可达的增量聚类算法——ICADD算法，由于该算法是非批量聚类方式，实验表明虽然聚类效果比较好，但是效率较低；（2）根据BIRCH 中聚类特征的概念，簇特征设计实现了 Abstract With rapid development of network technology and database technology, the amount of information grows rapidly in various fields and the data scale becomes more and more huge. Besides the increasing demand for real-time processing leads to the emergence of massive dynamic data. Most of traditional clustering algorithms consume large amounts of time and space resources and they are of weaknesses in effectiveness and scalability. In this case, the emergence of incremental clustering algorithms can slove the problems effectively. In order to achieve dynamic, incremental clustering, this paper analyzes Clustering Algorithm Based on Density and Density-reachable （CADD） at first, and then makes some improvements. The improvements include three aspects: 1. Set signs for ensity-reachable members. Aim is to improve efficiency. 2. Improve method for calculating radius and density which avoid double counting. 3. Achieve visualization to assess clustering results effectivly. Experimental results show that improved CADD algorithm reduces the complexity. Based on improved CADD algorithm, the paper focuses on the following two aspects：（1）proposes Incremental Clustering Algorithm Based on Density and Density-reachable （ICADD） according to the characteristics of CADD algorithm. The algorithm uses non-batch mode. It is less efficient. （2） proposes Incremental Clustering Algorithm Based on Subcluster Feature （ICSCF） the notion of clustering featu