Research On 5G Telecom Customer Prediction Based On Data Mining

In the development of the information trend,as the leader of the 5G communication era in China,we ushered in the 5G era.With the continuous development of the communications in-dustry,With the development of information technology,we have ushered in the wave of the 5g era,and China is also a leader in the 5g communications era.In the process of continuous devel-opment of China's communications industry,the three major operators of China Mobile,China Unicom and China Telecom are doing everything possible to compete for user resources in the existing market,and the competition is fierce.Which operator can obtain user needs faster,more accurately,and more effectively based on user information,and which operator can win business opportunities and occupy a dominant position in the market.For telecom operators,data mining based on consumer behavior helps operators better carry out 5G telecom customer identification and other services,so as to promote 5G at the product,terminal,service,and network levels,and discover potential 5G users in a timely manner.Telecommunications user data generally includes basic information about the operator's infrastructure construction and basic behavior information of telecommunications users.The data used in this article is real data provided by a certain website.The data used in this article comes from the real data of telecom operators provided by a certain platform,which includes basic information about the infrastructure construction of the operators and basic behavior infor-mation of telecom users.The prediction of 5G telecom users is actually a two-class prediction problem.The process of constructing a prediction model mainly includes data preprocessing,feature engineering,and model training and evaluation.First,perform exploratory analysis and preprocessing of the data,and construct derivative feature variables,and then perform feature se-lection based on the Light GBM algorithm,and screen out 32 feature variables with high feature importance.Due to the extremely unbalanced data,the SMOTE algorithm was used to balance the data set,and then the Logistics regression algorithm,AdaBoost algorithm,XGBoost algo-rithm,Light GBM algorithm and CatBoost algorithm were used to construct the 5G telecom user prediction model,and based on the accuracy of the model on the test set Comprehensive com-parison of multiple model evaluation indicators such as degree,recall,precision,F1 value,AUC value and KS value.The results show that based on the data set in this article,the Light GBM model,XGBoost model and CatBoost model are better than the AdaBoost model and Logistic regression model in terms of comprehensive evaluation index F1 value and AUC value,indicat-ing the classification accuracy of these three models Higher and better model performance,able to distinguish 5G telecom users from non-5G telecom users.According to the analysis of the prediction model,factors such as the proportion of 5G users in cities,the construction of base stations and package types of operators,and the user's billing income have a great positive ef-fect on whether telecom users can be transformed into 5G telecom users.Operators can predict telecommunications users based on the prediction model,and identify the needs of users in a timely manner,and provide a basis for decision-making on the transformation of users into 5G telecommunications users.