一种提高DBSCAN聚类算法质量的新方法  

New method to improve DBSCAN clustering algorithm quality

作　　者：冯少荣[1] 肖文俊[1] 

机构地区：[1]华南理工大学计算机科学与工程学院

出　　处：《西安电子科技大学学报》2008年第3期523-529,共7页Journal of Xidian University

基　　金：国家自然科学基金资助(50474033)

摘　　要：针对基于密度带有"噪声"的空间聚类应用(DBSCAN)聚类算法存在的3个主要问题:输入参数敏感、对内存要求高、数据分布不均匀时影响聚类效果,提出了一种基于遗传方法的DBSCAN算法改进方案数据分区中使用遗传思想的DBSCAN算法(DPDGA)来提高聚类质量.利用遗传算法改进K-means算法来获取初始聚类中心;对数据进行划分,在此基础上对划分的每一部分使用DBSCAN算法进行聚类;合并聚类的结果.仿真实验表明,新方法较好解决了传统DBSCAN聚类算法存在的问题,在聚类效率和聚类效果方面均优于传统DBSCAN聚类算法.There are three problems along with the Density Based Spatial Clustering of Applications with Noise（DBSCAN） Clustering Algorithm： input sensitivity, desire for too much memory space and the effect of nonuniform data. To solve these problems, a fast Data Partition DBSCAN using Genetic Algorithm（DPDGA） Algorithm is developed which considerably improves the cluster quality. First, the Genetic Algorithm is used to improve the K-means Algorithm to get the initial clustering center. Second, data is partitioned and the DBSCAN Algorithm is applied to cluster partitions. Finally, all clustered result sets are merged. Simulation experiments indicate that the DPDGA Algorithm works well to solve these problems and that both the efficiency and the cluster quality are better than those of the original DBSCAN Algorithm.

关 键 词：聚类算法 遗传算法 数据划分 密度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]