基于邻域粒化的混合信息系统动态规则提取  

Dynamic rule induction for hybrid information systems based on neighborhood granulation

作　　者：程昳 刘勇[3] CHENG Yi;LIU Yong(College of Computer Science,Sichuan University,Chengdu 610000;Department of Information and Engineering,Sichuan College of Architectural Technology,Chengdu 610000;Department of Equipment Engineering,Sichuan College of Architectural Technology,Deyang 618000,China)

机构地区：[1]四川大学计算机学院,四川成都610000 [2]四川建筑职业技术学院信息工程系,四川成都610000 [3]四川建筑职业技术学院设备工程系,四川德阳618000

出　　处：《计算机工程与科学》2019年第7期1236-1243,共8页Computer Engineering & Science

基　　金：国家自然科学基金(61071162)

摘　　要：现有的混合信息系统知识发现模型涵盖的数据类型大多为符号型、数值型条件属性及符号型决策属性,且大多数模型的关注点是属性约简或特征选择,针对规则提取的研究相对较少。针对涵盖更多数据类型的混合信息系统构建一个动态规则提取模型。首先修正了现有的属性值距离的计算公式,对错层型属性值的距离给出了一种定义形式,从而定义了一个新的混合距离。其次提出了针对数值型决策属性诱导决策类的3种方法。其后构造了广义邻域粗糙集模型,提出了动态粒度下的上下近似及规则提取算法,构建了基于邻域粒化的动态规则提取模型。该模型可用于具有以下特点的信息系统的规则提取:(1)条件属性集可包括单层符号型、错层符号型、数值型、区间型、集值型、未知型等;(2)决策属性集可包括符号型、数值型。利用UCI数据库中的数据集进行了对比实验,分类精度表明了规则提取算法的有效性。The data types of existing knowledge discovery models of hybrid information systems are mostly symbolic,numerical conditional and symbolic decision attributes.Most of the models focus on attribute reduction or feature selection,but research on rule extraction is relatively few.We construct a dynamic rule induction model for hybrid information systems covering more data types.Firstly,the existing formulas for calculating value differences of different types of attributes are modified,and a definition of the distance of cross-level symbolic values is given,thus a new mixed distance is defined.Secondly,we propose three methods to induce the decision class for numerical decision attributes.Then,we propose a generalized neighborhood rough set model based on neighborhood granulation,and the lower and upper approximations of an arbitrary subset under dynamic granulation are presented,which underlies a foundation for the construction of a dynamic rule induction algorithm.The model can be used to extract rules from the information systems with the following features,namely:(1)condition attribute set includes single-level symbolic,cross-level symbolic,numeric,interval-valued,set-valued and missing data;(2)decision attribute set can include symbolic and numeric data.The rule induction algorithm is evaluated on several data sets from the UC Irvine Machine Learning Repository.Experimental results show that the algorithm can achieve good classification performance.

关 键 词：规则提取 混合信息系统 粒度 邻域 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]