An Intelligent Package Matching Model For Users In The Telecom Industry Based On Machine Learning

With the rapid development of 5G,major telecom operators have begun to vigorously promote 5G packages.In order to retain the original 4G package users,telecom operators need to provide suitable 5G package services for different telecom users.Therefore,if each operator wants to If you want to stand out in this war,you need to launch a 5G package business suitable for users in order to gain an advantage in the next business war.The data used in this article comes from the telecommunications user package data released by the China Computer Society and China Unicom Research Institute in March 2020.There are about 750,000 data,which are divided into 11 categories,and there is a serious category imbalance between the data.This article establishes an integrated machine learning model to provide telecom users with package services that are suitable for them.This article first introduces the theories,advantages and disadvantages of several machine learning models.Next,the data is preprocessed accordingly,and the missing values are first performed For operations such as deletion and interpolation,the interpolation mainly uses KNN interpolation.According to the original data characteristics,4 new features are built,and then the dimensionality of the data features is reduced,and the original data attributes are changed by sklearn's Select KBest Make a selection,reduce the dimensionality of the data to18 dimensions,and perform modeling processing on the selected data.In terms of model establishment,a single machine learning model and an integrated machine learning model were established respectively,and the single machine learning models created were KNN,logistic regression,and naive Bayes.The integrated machine model mainly builds the following models: XgBoost,GBDT,AdaBoost,random forest model.Among them,the basic classifier of AdaBoost is a decision tree model.And through the accuracy and F1 value to evaluate the model,and finally found that in a single machine learning model,the KNN model performed the best,and its accuracy and F1 value were 0.70 and 0.69,respectively.For integrated machine learning models,the XgBoost model performs best,with its quasi-deleting rate and F1 value of 0.90 and 0.90,respectively.Among them,the worst-performing model is the AdaBoost model,whose quasi-determination rate and F1 value are 0.83 and 0.83 respectively.It can be seen that integrated machine learning is more suitable for the package matching problem in the telecommunications industry.