Identification Of Potential Users Of 5G Packages Based On Statistical Learning Under The Background Of Big Dat

With the rapid development of the fifth generation mobile communication technology(Refered to as 5G),5G has been officially launched commercially in China,and the number of 5G end-users is growing at a rapid pace.At present,the competition among telecom operators in the 5G field has entered a white-hot stage,and the variety of telecom packages has become more and more abundant.The diversification of telecom packages has led to the exposure of the drawbacks of the traditional crude marketing strategy and the increasing refinement of users’ demands for packages.In addition,with the improvement of data collection and storage capacity,telecom operators have large-scale subscriber information data.In this context,based on the mobile subscriber information data of Chongqing area for three months in 2020,this paper investigates the identification of potential 5G users and the personalized recommendation of telecom packages,specifically.1、The identification model of 5G potential users is studied.Firstly,the preprocessed data were selected based on penalty variables,and the variables that have significant impact on the identification of 5G potential users were obtained.Secondly,the correlation analysis between the filtered variables and whether they are 5G users was conducted.Finally,a 5G potential user identification model based on logistic regression and random forest algorithm was established and evaluated in detail.After analysis,the following conclusions are obtained:(1)Random forest has advantages in recognition effect,with higher accuracy,recall,F1-score and AUC values than logistic regression;(2)In general,the accuracy of the random forest model after parameter tuning is 77.39%,which has the best recognition effect.2、The problem of personalized recommendation of telecom packages is studied.Firstly,the logistic regression modeling was conducted based on the service type of 5G package and user’s basic information data and behavioral information data,and the significance of each index was tested by wald test.Secondly,a collaborative filtering recommendation algorithm based on K-means clustering was proposed.The analysis results of this paper show that the accuracy of the collaborative filtering recommendation algorithm based on cluster optimization is 70.63%,which is 14.92%higher than the accuracy of the traditional collaborative filtering recommendation algorithm.In addition,the clustering-based collaborative filtering recommendation algorithm also saves computation time significantly.The new personalized recommendation algorithm provides a new method for operators to promote telecom packages.