基于邻域粒化的多数据源高投票决策规则的挖掘  

Mining high-voting decision rule based on neighborhood granulation in multiple data sources

作　　者：陈辉皇 林耀进 林国平 唐莉 

机构地区：[1]闽南师范大学计算机学院,漳州363000 [2]数据科学与智能应用福建省高等学校重点实验室,漳州363000 [3]闽南师范大学数学与统计学院,漳州363000

出　　处：《南京大学学报（自然科学版）》2017年第6期1063-1071,共9页Journal of Nanjing University（Natural Science）

基　　金：国家自然科学基金(61672272;61303131;61603173);福建省高校新世纪优秀人才支持计划

摘　　要：多数据源高投票决策规则挖掘是指从多数据源中挖掘存在大部分数据源且具有重要意义的决策规则,此类规则在银行理财产品营销、市场营销、疾病诊断等领域中具有指导性作用.利用样本邻域粒化来构建决策规则的表现形式,在此基础上定义了覆盖度、投票数等多种决策规则的度量指标,用以挖掘满足这些度量指标的高投票决策规则.实验结果验证了所提算法挖掘多源决策信息系统中的高投票决策规则挖掘的有效性.High-voting decision rule mining of multiple data sources is denoted that mining decision rules from multiple data sources,and these decision rules exist in most of multiple data sources and are significant.In many real-world applications,this kind of decision rule has instructive role,such as bank financial products marketing,marketing management,and disease diagnosis.Therefore,the purpose of this work is to discover non-trivial,interesting,and interpretable high-voting decision rules from multiple data sources.Firstly,a formal presentation of decision rule via the sample＇s neighborhood granulation is constructed.Then,some metrics about decision rule are defined,such as cover degree,and vote rating.These metrics reflect the significance and interesting degree of decision rule from different views.Finally,the concept of high-voting decision rule is defined according to the metrics of decision rule.Experimental results demonstrate the effectiveness of the proposed algorithm,which is used to mine high-voting decision rules from multiple data sources.

关 键 词：邻域粒化 多数据源 高投票决策规则 分类间隔 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]