基于观测数据偏差修正的DBSCAN改进算法  

DBSCAN improvement algorithm of deviation amendment based on observed data

作　　者：曹亮[1] 舒红平[1] 刘魁[1] 

机构地区：[1]成都信息工程学院计算机系,四川成都610225

出　　处：《计算机工程与设计》2009年第13期3160-3162,共3页Computer Engineering and Design

基　　金：四川省青年科技基金项目(06ZQ026-054)

摘　　要：在经典DBSCAN聚类算法中距离度量直接影响到聚类的结果,为了使观测数据的偏差不影响到距离度量,在研究观测数据预处理过程及经典DBSCAN聚类算法基础上,提出了一种DBSCAN距离度量的改进算法,指出了观测数据偏差修正的基本思想及改进算法的处理流程。仿真结果表明,改进算法能很好地修正观测数据的偏差,最终得到较好的聚类结果。Distance measure direct effect to clustering result in classics DBSCAN cluster-algorithm, so the least affect to distance measure for the deviation of observed data. Based on the research of observed data pretreatment process and classics DBSCAN cluster-algorithm, an algorithm of DBSCAN distance metrical improvement is brought forward, and the fundamental thought of observed data amendment and the treatment process algorithm is pointed out. Simulated result show that improvement algorithm to be able to great amend the observed data deviation, get preferably clustering result ultimately.

关 键 词：聚类算法 预处理 观测数据 偏差 距离度量 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]