Research On Mobile Users Of Telecom Operators Based On Data Mining

In the wave of big data development,China's telecommunication enterprises have become data pipelines because of constant impact of OTT(Tencent,Ali,etc.)enterprises.It is not only opportunities but also challenges for telecommunication operators.With the increasing saturation of Internet users in China,the growth rate of telecom customer has changed into low-speed from high-speed.Under the background of the national implementation of "number portability" and the new situation of international telecommunication operators' access to the domestic market,it is crucial that how to maintain the stock customers and develop high-quality customers for enterprises.Compared with other enterprises,telecommunication operators have the upper hand for the amount of basic data,especially in the background of real-name customer system,which improves the authenticity of enterprise customer data.Data mining with the advantage of its own data resources can help telecommunication operators maintain the stock customers,improve the development quality of new customers,improve the competitiveness of enterprises,and then realize the digital transformation.A series of basic data sets provided by a telecommunication operator are used to do the following work and research in this paper:(1)Based on Hadoop technology and the basic knowledge of data mining,this paper thoroughly studied the establishment of linear regression analysis model and parameter estimation method.The process of how to creating decision tree and select the best attribute and common pruning methods are researched.(2)Using the time window method,the short message and voice communication records between users are quantified.In this paper,a mobile user interaction index model based on multiple linear regression is proposed,and the parameters in the model are calculated by the least-square method.This model can complete the calculation without complex operation,and can be widely used in the production environment of enterprises.It provides an important basis for the follow-up work and research.(3)A model of fake mobile user recognition based on decision tree is constructed by using the calculation results of the stable association coefficient model and the basic data set of operators.Due to the imbalance of the sample dataset,the traditional C4.5 algorithm aims at the lowest error rate,which leads to the classification result of the model tending to have more classes and can not satisfy the requirements of the fake user recognition model,so this paper makes two optimizations: Firstly,apriori algorithm based on association rules mines and selects rules with high support and confidence,then finds out which important attributes affect fake user recognition,retains key rules,and applies the selected rules as new attributes to decision tree according to specific conditions.Secondly,traditional decision tree algorithms mark leaf nodes and model pruning based on the minimum error rate,which can not fully satisfy the requirements of fake user recognition.Therefore,this paper uses a cost-sensitive method,which classifies the model based on the minimum error.The experimental results show that the optimized algorithm improves the recall rate of the model and the classification effect is greatly improved.