| In recent years,with the gradual saturation of the market share of the operator industry and the gradual slowdown of the scale growth,the major operators have gradually started the competition among the stock users.How to maintain the better service provided by users in the network,how to attract users from other operators to change operators through port number transfer,etc.This kind of competition also promotes the diversification of package types in the market,however,it also leads to the inability of users to make on-demand choices in front of rich products.Therefore,it is an important direction for the development of telecom industry to analyze the consumption information,calls and traffic usage habits of users and recommend suitable intelligent packages to users.In this study,based on the historical data of more than 740,000 users’consumption behaviors published by the telecommunications industry,combined with the methods of data mining and machine learning,how to complete the intelligent package recommendation for users is studied.In this study,firstly,the data is preprocessed,the preliminary exploration of the data is completed,and the missing values are analyzed and processed,which lays the foundation for the subsequent feature analysis and model building.In the stage of feature analysis,according to the dimension of sample data,the feature analysis is completed from the aspects of user’s call duration,traffic usage,consumption situation,complaints and user’s basic information,and new features are constructed in combination with business situation,which provides convenience for subsequent model construction and prediction.In the aspect of model construction,based on four machine learning models of KNN,random forest,GBDT and XGBoost,the training and prediction of the models were completed,and the effects of several models in package recommendation were evaluated,and the model algorithm with high applicability in user package recommendation was obtained.The research shows that the bill consumption characteristics and traffic characteristics of users are important characteristic indicators for users to choose packages,and the types of packages and the duration of contracts are also concerned by users when choosing packages,while the types of contracts,gender and other value-added information of users are relatively unimportant.In the subsequent package design or optimization,the telecom industry can make timely adjustments according to the parts that users pay attention to.Among the four machine learning algorithms,XGBoost model has the best effect,with the accuracy rate of 0.92,the recall rate of 0.91 and the F1 value of 0.92.The results show that XGBoost model has a good prediction result,and has a good application effect and prospect in intelligent package recommendation for users. |
