银行决策支持系统中数据挖掘的研究与实现

【作者】 周四新；

【导师】 陈松乔；

【作者基本信息】 中南大学 ， 计算机应用技术， 2004， 硕士

【摘要】 在国外银行即将进入中国市场之前，提高国有银行的决策质量以增强其市场竞争力，具有极其重要的意义。基于这种情况，我们和中国农业银行湖南省分行合作开发研制了农行决策支持系统项目。 系统采用基于B／S模式的四层体系结构，界面层采用Jsp技术访问Web服务器，中间层应用服务器存储以JavaBean形式表示的业务逻辑，底层是用Sybase ASE 12.0构造的逻辑数据仓库。 论文第一章介绍了决策支持系统、数据仓库和数据挖掘的研究现状；第二章给出系统的总体结构；第三章着重阐述了逻辑数据仓库的构造；第四章对系统所采用的挖掘构件做了详尽的描述；第五章是我在研发过程中的工作总结。其中，第三章和第四章是论文的重点。 对于数据仓库的设计，我们从概念模型设计开始，首先分析系统的需求，从而确定数据仓库的主题，并且采用星型模式进行建模。然后是逻辑模型设计，对概念模型进行进一步的细化，对每一个主题域确定其维表和事实表的表结构。最后进行物理模型设计，针对系统实际的硬件环境，确定数据的存储结构，数据的索引策略，以及数据的存放位置。 系统针对银行的具体业务情况，设计了三个挖掘操作构件。分类构件所采用的是基于对SPRINT算法进行改进的NS判定树分类算法，其可扩展性和挖掘时间都得到了很好的改善。序列模式挖掘构件、聚类构件分别采用的是AprioriSome算法和CLIQUE算法。更多还原

【Abstract】 It is significant to improve the quality of decision making of state banks and thus enhance their competency against overseas banks that are entering Chinese market.For this reason, the project "Agricultural Bank’s Decision Support System(ABDSS)" has been developed in the cooperation with Agricultural Bank of Hunan province.A four-level structure based on B/S form is employed in the system, that is, in the interface level, the web server is accessed using the JSP technique; in the middle levels, the business logics are stored in the server by means of Java Beans; and the bottom level is a data warehouse developed by Sybase ASE 12.0.In the first chapter of this thesis, current advances in the research of decision support system, data warehouse and data mining are introduced. The second chapter proposes the system structure and design ideas. The third chapter discusses mainly about the development of the logic data warehouse. The fourth chapter gives a detailed description on the data mining component applied in this system. And in the fifth chapter, a conclusion is drawn about the work done in this research. The third and fourth chapters form the main part of the whole thesis.Regarding the data warehouse design, it starts from the design of conceptual model. Firstly, the system requirements must be analyzed thus the subjects of the data warehouse can be determined, in which the star schema is used for the modeling work. Next, the logic model design is carried out. In this process, the logic model should be further developed and improved, that is, the structure of dimension tables and fact tables for each subject will be determined. Finally, the physical model design is performed, in which the data storage and index strategy are considered according to the actual hardware conditions.Aiming at specific business situations of the bank concerned, three mining operation components are designed. For the classification component, a decision tree classification algorithm named NS is used for application, which derives from the SPRINT algorithm and has beenimproved, so that it can get a better performance on both the extending ability and mining time. Concerning the sequence pattern mining component and clustering component, the AprioriSome algorithm and CLIQUE algorithm are employed respectively for their development.更多还原