Realize The Financial Customer Analysis Using Data Warehouse

This thesis researches on how to build the personal credit data warehouse to make a customer analysis on which the credit decision of the commercial bank will be based, with the background of data warehouse, data mining and their solutions being grown up. Personal credit data warehouse contains the data of customers, loan types, accounts, branches and operators which will be an auxiliary tool for decision-making from all aspects. The warehouse can also be used to analyze the customer group and their behaviors, or find the key credit products and regions, developing directions and problems. It provides an accurate base for credit business managing, risk control, customer relation managing and market promotion which enables the business department of the commercial bank to make a better decision to meet the market change, follow the developing direction and improve competing capability. This system provides kinds of credit statistic reports and data with two or three dimensional graphs, pie charts and bars. Some information in common use can be provided in fixed formats with the help of systematic models. The users of different levels can also customize the statistic reports with their different privileges. The system contains the following parts. 1.Data extract, transform and integration(ETL).There are three ways for the ETI process: using data copy mechanism, using tools offered by product-maker, programming a specific application. In this system the last way is chosen in light of the data distribution, network width and specific application. The running process is automatically controlled by the system. 2.Data storage and managing.Because of very large size of data in data storage and managing (often tens of GB, sometimes TB) China Construction Bank adopted a large-scale relational database with some components customized for parallel processing and decision search. 3.Data warehouse design and modeling.The principles of data warehouse are different from the traditional relational database. To meet the multi-dimensional needs, a star model and a snowflake model are used to build a multi-dimension model and provide OLAP analysis. 4.Data analysis and display.Using browser to implement OLAP analysis and display can have uniform management and flexible distribution, which accords with the enterprise Intranet managing mode. The system adopts the multi-level client/server structure which includes a data warehouse storage center based on dynamic servers with the expanded decision support and parallel processing components, an online analyzing server composed of OLAP server, credit managing application servers, web servers and Internet/Intranet browsers for clients. 5.Data mining.Data mining is an advanced method for researching and modeling on large-scale data. Data mining has a series of effective means: SEMMA-S(data sampling), E(data exploring), M (analysis and pretreatment), M(data adjusting, modeling and knowledge discovery), A (comprehensive explanation and evaluation of the model and knowledge). Data sampling Data may be random or in levels, alternative or unalternative. But the data in data sampling should be unalternative for a credit counting purpose. The data warehouse can have hundreds of GB so if you want to know all the rules of the enterprise you must select proper subsets in the whole data set. Data exploring, analysis and pretreatment It should have some statistic techniques such as hierarchical cluster analysis, factor analysis and correspondence analysis in a GUI manner. Data adjusting It adjusts the parameters of the analysis, classifies and integrates the parameters into needed subsets. Modeling and knowledge discovery It contains comprehensive statistic means such as artificial nerve network, time series analysis, decision-tree( CHAID and CART), standard statistic methods(hierarchical cluster analysis, discriminant analysis, logistic regression, survival analysis etc.) Comprehensive explanation and evaluation for the model and knowledge It makes a comprehensive evaluation for the model and selects the best one. It will repeat the data mining process if a new question emerges. The system has some traits:The multi-dimension principles are embedded in the design and modeling of the data warehouse, and a specific business model is made for CCB project management. The data warehouse applications and traditional managing applications are integrated in a multi-level software logic structure design. A high-powered relational database can not only improve data management and maintenance but also protect the value and connect with the business system seamlessly. A specific data integration solution including data extract, cleansing and collecting is provided for the applications.