Study Of High-value Mobile User Prediction Based On Telecom Big Data Mining

Along with the rapid development and popularization of the “internet+” business model,the competition of the mobile communication market is fierce day by day.The marketing operation of high-value mobile user has become an important avenue for telecom operators to increase profits and market competitiveness.Based on the telecom big data mining,a decision rule of high-value mobile user is determined with the in-depth analysis of massive mobile users in the consumption and loss,and the corresponding prediction model is also established,which has important research significance to increase the income of enterprises and improve market competitiveness.In this paper,by combining with the Teradata data warehouse,data mining tool R and CRISP-DM methodology,the high-value mobile user data which possess the characteristics of high contribution and low loss is extracted from a telecommunication company.After data preprocessing,modeling and model evaluation,the prediction model based on random forest algorithm is selected for further study.Finally,aiming at the defects of the random forest which is easy to be biased towards the majority of the binary classification on the high dimensional imbalanced data set,this subject proposes a method of effective feature selection to optimize the prediction model.With balanced training sets extracted from an initial imbalanced data set using under-sampling,a feature selection strategy based on Pearson correlation analysis and random forest method assessing the feature's importance is applied and the best feature subset is selected by embedding weighted and voting mechanism in the ensemble learning method.The final prediction model is built by random forest algorithm.The experimental results show that this optimization method can effectively reduce the dimension of feature set,and enhance the prediction accuracy and generalization ability of the model for high-value mobile user,so as to provide better service for enterprise decision.