运用自适应粒化模型的一种数据分类方法  

Data Classification Method Using Adaptive Granulation Models

作　　者：蒙祖强[1] 沈亮亮[1] 甘秋玲[1] 覃华[1] 

机构地区：[1]广西大学计算机与电子信息学院,南宁530004

出　　处：《小型微型计算机系统》2018年第3期417-424,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61363027;61462006)资助;广西自然科学基金项目(2015GXNSFAA139292)资助

摘　　要：数据分类是一种重要的数据分析技术,但数据分类方法大多涉及诸多人为参数优化问题.在现今复杂数据环境下,事先获取数据的有关先验知识是一件比较难的事情,这给解决参数优化问题带来困难,限制了数据分类方法的应用范围.因此,如何构造无需参数优化的自适应分类方法是数据分类研究领域中面临的一项重要课题.本文构造一种自适应粒化模型并据此提出一种自适应数据分类方法.该方法以数据对象为中心,充分利用数据之间蕴含的相容关系,通过相容半径的自动优化来构造自适应粒化模型,然后基于此模型,通过数据对象的约简和对象规则的简化来构造分类器.整个分类方法完全是由数据驱动的,不需要任何先验知识和手工参数优化,实现了对数据的自适应分类.最后,通过实验验证了所提出方法的有效性.Data classification is an important data analysis technology, but most of data classification methods need to deal with issues of manual parameter optimization. Today, it is still difficult to acquire related prior knowledge from complex data environment, which causes difficulties for solving the problem of parameter optimization and limits the scope of its applications. Therefore, how to design adaptive classification methods without parameter optimization is an important subject in the field of data classification. In this paper, an data classification method was proposed based on adaptive granulation models. The method centered on data objects and took full advantage of tolerance relations between data values, and then constructed an adaptive granulation model by automatically optimizing tolerance radius. Based on the constructed granulation model, a classifier using for data classification was contructed by reducing data objects and reducing rule set. The proposed data classification method is data-driven completely, without any prior knowledge and man- ual parameter optimization. Experimental results also show that the proposed algorithms are effective.

关 键 词：数据分类 自适应 分类器 约简 粒化 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]