一种新的位置不确定性聚类算法UCNDBSCAN  

Clustering Method for Position Uncertain Data Based on Connection Number-ucndbscan

作　　者：王骏 黄德才[1] WANG Jun;HUANG De-cai(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 3100230 China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023

出　　处：《小型微型计算机系统》2018年第8期1633-1640,共8页Journal of Chinese Computer Systems

基　　金：水利部公益性行业科研专项项目(201401044)资助

摘　　要：位置不确定性数据的聚类是一个新的不确定性数据聚类问题.目前对于这一类问题的聚类算法主要以划分聚类为主,而划分聚类有着无法区分任意形状簇和无法分离离群点等缺点;已有的一些基于密度的聚类算法,存在单单考虑对象间距离的均值,忽略距离变化范围,参数敏感性大和计算复杂度高等缺点.鉴于此,提出一种基于联系数的位置不确定数据密度聚类算法-UCNDBSCAN.该算法用联系数巧妙地表示不确定性对象,并专门定义了对象间的联系距离,运用联系数态势值理论定义新的对象间距离衡量标准,克服了现有算法的不足.仿真实验表明,UCNDBSCAN具有聚类精度高、参数敏感度低、计算复杂度低、实用性强的特点.Clustering for position uncertain data is a new problem of uncertain data clustering. Right now,the main solution to this new problem of clustering is partition clustering. However,the partition clustering has disadvantage of disability of distinguishing any shape of cluster and disability of identifying outliers. Presently, the density clustering method for the uncertain data has disadvantage of only considering the average of distance between objects ignoring the variation range of distance and disadvantage of high demand of input- ting parameters and disadvantage of high cost of computation operation. Therefore,a new uncertain data density clustering method-UC- NDBSCAN is put forward. This method uses connection number as model of uncertain object and defines connection distance for the distance between two objects and use Shi value to define the new standard of measuring distance among uncertain objects, which over- come the disadvantages existed in the solutions above. The experiment illustrates that UCNDBSCAN has high precision of clustering and low dependency on parameters and low complexity and easy use.

关 键 词：不确定性数据 密度 联系数 聚类 数据挖掘 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]