Research On 5G Package Potential Customer Identification Based On E-LightGBM Algorithm

With the rapid development of 5G mobile communication technology,the wave of5 G era has surged forward.In 2021,the state clearly proposed to accelerate the largescale deployment of 5G network in the "fourteenth five year plan" and the 2035 longterm goal outline,and the 5G user penetration rate has increased to 56%.Obviously,as a carrier providing 5G network services,the 5G package business of operators must be one of the important starting points to promote the accelerated development of 5G.At the same time,its rapid popularization will also be one of the keys for operators to gain competitive advantage in the fierce 5G market.However,from the current 5G package marketing and promotion process of major operators,there are still some potential problems,such as difficult positioning of user needs,unclear marketing users Package recommendation is not targeted,etc.Therefore,how to mine and analyze users' demand changes and consumption changes with existing technical tools? How to effectively and accurately identify potential users of 5G package? And further developing these potential users into real 5G package users are issues that major operators need to focus on when promoting 5G package services.In this context,from the perspective of practical application,based on the massive data accumulated in the big data platform of operators,this paper analyzes the behavioral differences or common characteristics between 5G package users and non 5G package users,and uses data mining technology to build a 5G package potential customer identification model to predict potential demand users with 5G package handling tendency in a specific time in the future,so as to reduce the blindness of 5G package marketing,Improve the success rate of marketing recommendation.Based on the full text,the main contents of the study are as follows:Firstly,an LMS algorithm(LightGBM-MIC-SBS)based on feature factors is proposed for feature selection to express the contribution degree to the model.The algorithm comprehensively considers the importance of features to the model and any correlation between features,that is,the feature influence factor defined by the combination of the feature importance output from the LightGBM model and the maximum information coefficient MIC is used as the feature screening evaluation index,Then the sequential backward search method SBS is used to select features.The research shows that LMS algorithm can effectively screen the features with high importance and low redundancy,has the characteristics of small human intervention and fast screening speed,and can play a role in the actual feature selection.Second,in order to improve the model's ability to identify potential users of 5G packages,this paper constructs an Easyensemble-LightGBM model(E-LightGBM for short)based on the combination of under sampling technology and integrated learning technology.This model improves the Easyensemble method in three aspects: the base classifier,the balance ratio of positive and negative samples,and the result output integration strategy.Through the horizontal comparison of different data balancing methods combined with LightGBM model and the vertical comparison with other different models,it is found that E-LightGBM model shows better results in all comparison models.After verification with future real data,it is further proved that the model has certain effectiveness and efficiency,and can play a role in the identification of potential customers in the actual 5G package.