The Research Of Enterprise Decision Support System Based On Data Warehouse And Data Mining

Because the database result in inadequate information for decision owing to only the general processing and collecting of the initial data and the model-base is difficult to adapt to the dynamic quality and complex quality of the decision, the traditional decision support system base on database and model-base that the two are independent of the design and lack of internal unity. Nowadays, with the development of computer technology, the new decision support system based on data warehouse, online analytical processing and data mining has become a progress trend. In order to keep up with the trend, this dissertation takes the retail enterprise sales forecasting as research object, and studies data warehouse construction, multi-dimensional data set, online analytical processing and data mining algorithms of sales forecasting in the new decision support system. Moreover, a three layer C/S-pattern retail enterprise sales forecasting decision support system is designed to validate the proposed technologies. The main contents and conclusions of the research include:(1) Aiming at the requirement of enterprise information building and market competition, on the basis of the analysis for the deficient in the traditional decision support system, the new decision support system frame based on data warehouse and data mining is proposed and analyzed.(2) In order to integrate the business data of distributing in enterprise different networks site and separate information analysis environment from operating environment, the data warehouse of decision theme for sales forecasting is constructed according to the demand analysis, established logical model and relevant fact table and dimension tables, schemed out the data warehouse physical model in SQL Server 2000, and designed DTS package of data extraction, conversion, loading and automatic update. It will offer all types and trusty data for decision effectively.(3) For effectively analyzing data in data warehouse, and helping enterprise to gain information of the operation and the market demand, this paper utilizes OLAP technology to build the multi-dimensional data set of analysis, and researches it using slicing, dicing, drilling and rotating tools, make use of multi-dimensional query language MDX to enquire data set and find the useful information. It is put forward the three solutions that show OLAP analysis data.(4) For the purpose of predicting sales quantities and sales profit progress trend in the future, this paper uses data mining technology to study them. According to sales forecasting character, the four linear regression analysis model which builds commodity sales impact factors (sales date, commodity kinds, commodity prices, Customers purchasing power) and sales quantities is brought forward, and validate the model using FoodMart enterprise sales data. Experiment results show that the mean absolute percentage error of this method is 66.77%, the model is inapplicable.(5) Because the regression analysis model only describes causality of all kind of economic variables with a static view and less considers the dynamic development deficiencies of the true economic activity, the time series AR prediction model is proposed. It is studied that AR model builds steps, AR sales forecasting model is established. Experiment results show that the mean absolute percentage error of this method is 18.3%, it is good prediction model and can go on sales forecasting.(6) Considering the defect that linear regression model and AR model all adopt linear statistical analysis, data mining sales forecasts based on BP neural network is put forward. The network structure of sales forecasting is built according to research BP neural network. The input of network are a certain commodity sales time, commodity prices, customer purchasing power, the time series of AR model, the output of network are a certain commodity sales quantities of the different place, then build the three layer network structure, and go on network training utilizing adaptive learning rate and additional momentum method. It makes sure historical data length of the higher forecast accuracy through experiments. Experiment results show that the mean absolute percentage error of this method is 15.13%, the forecast accuracy is higher than the AR model.(7) Based on the research of data warehouse, data mining and online analytical processing, developed the three layer C/S-pattern retail enterprise sales forecasting decision support system using modularization method. The system can realized the function of management data warehouse and multi-dimensional data set,OLAP multidimensional analysis, sales quantities and sales profit forecasting based on the AR and BP networks algorithm models. Building the AR and BP forecasting models can realize the relearning and training. The system is practicable and expansible. Using FoodMart enterprise sales data to testify the system, the results indicated that the system can help enterprise to constitute scientific and rational sales decision plan, and is feasible.