Telecom Client Loyalty Prediction Model Research And System Representation

With the development of data mining, experts attempt to solve problems in various areas by using this technology. The main operation mode for this kind of technology is to build module for existing research of data analysis, then attempt to discover the hidden rules among those massive data. Obviously that telecommunication is the career of owning massive data, all companies are facing unavoidable client missing with the saturation and strong competition in the market. In order to keep company's competitive ability, all service providers have to get new client by greatly increase the expense since the cost for company to get new client is five to ten times more expensive that maintain existing client, and the profit contribution of old loyal client is much higher than new clients. Therefore, how to find the low loyalty level client and change this situation is the priority issue that we need to solve. This makes the data mining technology important and critical to find the elements that will affect clients'loyalty.The objective of this research paper is to discovery describable symptoms for client by using data mining technology, and then build client prediction module based on that. According to this module, it should predict the level of clients'loyalty and print out spreadsheet that company can create appropriate solutions for them.(1)The algorithm of data mining. There are a few algorithms in data mining, like genetic algorithm, bayesian network algorithm and decision tree algorithm etc. Each algorithm has its own suitable situation. This research paper chooses the bayesian network algorithm as the algorithm for building prediction module since it advances on data processing with the situation that parts of the data is missing.(2)The build of client loyalty module. This process have five steps:business objective confirmation; data analysis; data preparation; module build and module evaluation. Each stage has different assignment. In stage to confirm business objective, the assignment is to predict the loyalty level of each client, and provide list for company manager to process. Data analysis stage's assignment is to familiar with existing client information and analysis those data. Those client data can be client registration data, client consumption data, client communication data and others. In data preparation stage, valuable information will be collected from existing client data, by analyzing those useful information we could find the key element that will affect client's loyalty. And then those elements can be processed to satisfy the module build requirement for bayesian network algorithm algorithm. This process contains the following four steps: data selection, data filter, property selection and domain change. In the stage to build module, the assignment is to confirm the network structure for client loyalty module as well as the condition probability spreadsheet on each network node. In stage module evaluation, the quality of the module will be tested.(3)The design for client loyalty prediction system is used for build a system that can satisfy client needs. First of all it analysis the client loyalty prediction system, gained the understanding of the requirement for the system ability and functionality, then design the detailed function for the system and introduce the main functionality. The development technology for this system is JSP, database SQL Server2000, connected to the system database through JDBC technology. According to the test, this system met the original objective that can be used to predict client loyalty.