Research On Early Warning Of Anonymous Telecom User Churn Based On Data Mining

With the advent of the 5G era,how to enhance the loyalty of the stock users and effectively expand the new user groups has become an important issue for the major operators to solve.Based on the telecom time series data,this paper studies the prediction method of customer churn,and then expands the prediction method of customer behavior.Through this means,operators can make targeted operational strategy adjustments to meet the user experience,while indirectly enhancing the corporate image of operators.This paper focuses on three difficulties,which are data dimensionality reduction,user churn prediction modeling and improving prediction accuracy,to assist operators to make strategy and provide data support for scientific decision-making.The main results are as follows:(1)Since the effect of other feature engineering methods is not obvious,a feature engineering method based on SVD and k means is proposed.The new method makes use of the monthly payment data derived from the billing system,SVD Algorithm and K-means Algorithm to find out the optimal data subset in the current situation from two aspects of data dimension reduction and user label construction.By comparison and analysis,we choose the singular value sort and the accumulated value as the verification standard,and prove that the data subset selected by the new method can contain most of the information of the original data set while reducing the dimension of the original data set.(2)In view of the importance of user behavior trends in the strategy-making of operators,a user churn prediction model based on extreme gradient promotion tree(XGBoost)and lightweight gradient promotion Algorithm(Light GBM)is proposed,the optimized data set is used to train the prediction model,and the new data generated by the business system is used to verify the model.The experimental results show that the fused model has better recall effect and F1 value.Compared with the single model,the F1 value is increased by 3.16% and 1.66% respectively.(3)Based on the feature engineering method and the fusion loss prediction model,the early warning prototype system of telecom customer loss is designed.Combining the software engineering system development process and the actual application scene,the tasks of the prototype system are disassembled,and the modules of data input,data analysis and data export are designed and implemented.To sum up,this paper aims at the characteristics of telecom users' data and the problems of forecast precision,improves the forecast effect from two aspects of feature selection and model construction,and provides a new solution for the loss forecast of telecom operators.