Research And Implementation Of Loss Of Customers Prediction System In Telecommunication Based On Data Mining Technique

Data accumulated are more and more with the rapiddevelopment of database technique and application of databasemanagement system. For the reason that important informationis veiled in the proliferating data, further data analysis isexpected by people so as to make full use of them. Thetraditional relational database is highly effective in input, queryand calculating but can not find the rules and predict the trendaccording to existing data. In general, the IT circle lacks methodto discover the veiled knowledge, which leads to the fact thatdata are vast but knowledge gained from them is less. Thus datain large database become tomb, which is not easy to be visitedany more. Decision-making is not based on data but depends ondecision-maker's intuition, for the reason that decision-makersare short of method to extract valued knowledge frommagnanimous data. In addition, as the expert system technique isconcerned, it asks users or experts to input information. In thistime and money consuming process, errors and warps often cannot be avoided. So people expect highly effective system thatfinds out valued knowledge in magnanimous data storage so asto help make right judgment for enterprises and researchinstitutions and change data tomb to data gold. In the forum ofthe eleventh International Artificial Intelligence Associatedmeeting held in Detroit, U.S.A. in August, 1989, KDD(Knowledge Discovery in Databases) first appeared. In variousacademic works and journals, data mining has been consideredas the thesaurus of KDD for the purpose of focusing on nucleusresearch and expression convenience. In the forth InternationalKDD and Data Mining Conference held in New York, U.S.A. in1998, not only scientists discussed the subject, but more thanthirty software companies displayed their data mining products,most of which are applied in countries in North America andEurope. In China, though long time research on data mining hasbeen on in many institutions, there is no case of successful datamining technique application. However, in banks andtelecommunication enterprises, there exists the need andprecondition to apply data mining technique due to the highdegree of informationization. The application intelecommunication field includes management of customerrelationship, analysis of customer humbug, analysis of customerloss, analysis of consumption model and analysis of marketpromotion.As China entered WTO, global market demands domesticpublic telecommunication operators try to catch up withadvanced foreign counterparts so as to meet the internationalcompetition. To retain a customer saves more cost than todevelop a new one. Besides, the intense competition in mobiledivision increases the cruelty. Competition for market sharecauses more and more customers leave the previous network.As data intensive industry, telecommunication enterpriseshave accumulated vast data in recent years as the result of rapiddevelopment. How to make use of these data to predict whichcustomers will leave in the near future and to lower the loss rateis the urgent need for telecommunication enterprises.This research aims at implementation of data miningtechnique and applies it to prediction system of customer loss intelecommunication field to improve telecommunicationenterprises' ability to compete so as to increase profit.Theories of data mining and relative arithmetic areintroduced and then on the basis of actual project andCRISP_DM (Cross-industry Standard Process for Data Mining)as framework, the design and implementation of predictionsystem are realized. The sequence of demonstration is businessunderstanding, data understanding, data preparation, modeling,choice of model and evaluation and development.mathematical statistic tool SPSS and data mining softwareClementine are used during the process.Without any wonder data mining technique is on the basis ofmagnanimous data, as far as which is concerned, data extract,organizing and filter needs much labor, which has been provedin similar database project. Apart from that, the work is moreimportant for Chinese telecommunication enterprises due to theirown characters, such as complex data origin and data impurity.On the basis of CRISP_DM model, this thesis introduces theseprocesses of data extracting, data filtering, data rebuilding andbuilds the Data Mark finally. This thesis also encloses storedprocedures and major structures of table.After elementary organization, we combine data statistic anddata mining knowledge by the following software: SPSS andClementine and apply it to actual project. We observe feathers ofdata by using SPSS and mathematical statistic knowledge tofamiliarize ourselves with data and operation and then data arefurther filtered to prepare for the model building.Using the Clementine tool and arithmetic of decision treeand artificial neural network, we build the model and evaluatethe model with the Analysis node of Clementine. After it weevaluate the model with actual data using the SPSS tool, thenapply the most appropriate arithmetic to practice. At last wedisplay the data by Business Object.During the data mining procedure, we will switch the abovesteps to adjust and observe data, build model, and re-just datauntil gain the result we want. The above steps will be repeatedfor several times, which is the feather of data mining project.Even the model is built, new problems, such as launch of newoperation and time, emerge. We need to rebuild model andre-adjust data, which leads to repeated data mining process.Data mining is the result of people's long-time research anddevelopment on database. Today, the mature technique with ofhigh performance relational database engine and broad dataintegration puts data mining technique into practice in databasefield. The technique has bright future in telecommunication field.What the thesis contributes is how to apply it to enterpriseseffectively, considering the present condition intelecommunication enterprises, to predict by using accuratemodel so as to let it take effect as soon as possible.