基于决策粗糙集的代价敏感知识获取方法及其应用研究

项目名称： 基于决策粗糙集的代价敏感知识获取方法及其应用研究

项目编号： No.71201076

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 管理科学与工程

项目作者： 李华雄

作者单位： 南京大学

项目金额： 22万元

中文摘要： 决策粗糙集理论是对经典粗糙集理论的拓展，由于引入了贝叶斯风险决策方法，其可用于获取数据中的代价敏感知识，目前正成为国际粗糙集研究领域的新亮点。本项目以信用风险评估问题为背景，以决策粗糙集为工具，针对决策者风险偏好差异以及误决策代价非平衡的特点，系统研究基于决策粗糙集的代价敏感知识获取方法及其应用。内容包括：（1）研究基于决策粗糙集的风险偏好建模及代价敏感决策边界的界定，为决策者提供个性化代价敏感决策支持；（2）研究决策粗糙集的代价敏感属性约简理论、约简算法及规则提取方法，以实现从海量数据中获取精简知识；（3）研究不完备数据基于决策粗糙集的代价敏感知识获取方法，实现决策粗糙集模型在不完备数据中的拓展，为决策者在非完整信息条件下的作出科学评估方案提供参考。在理论方法研究的同时，本项目将以银行信用评估的实际问题为研究对象，依据理论方法研究结果，开展应用研究，为科学合理评估信用风险提供决策支持。

中文关键词： 决策粗糙集；代价敏感；知识获取；属性约简；序贯决策

英文摘要： Decision-Theory Rough Set (DTRS), a new extension model based on classical Pawlak rough set, recently has aroused wide attention in the research area of rough set due to its cost-sensitive analysis capability via Baysian theory. Based on decision-theory rough set, the main research objective of this proposal is to establish a framework of cost-sensitive knowledge acquisition from the data resource for credit risk assessment problem, in which the decision makers have multi-view risk preferences and the misclassification costs are unbalanced. The main tasks of this project are to: (1) establish a mathematical model of risk preferences based on decision-theoretic rough set and present the decision boundary of the decision makers with different risk preferences, by which the personalized decision supports are presented; and (2) develop a new cost-sensitive attribute reduction theory which may preserve the misclassification cost of attribute sets, and present a search algorithm to find the proposed attribute reduction, and decision rules based on the reduced data are induced; and (3) investigate the cost-sensitive knowledge acquisition methods for incomplete data based on decision-theoretic rough set model, and present decision support for rational credit risk assessment in the case that the available information for

英文关键词： decision-theoretic rough set；cost-sensitive；knowledge acquisition；attribute reduction；sequential decision