| China is in a critical period of transition from 4G society to 5G society.China’s development strategy focuses on building smart cities and accelerating the in-depth integration of 5G user growth and urban development,and solving problems brought about by urbanization through informatization.The large-scale connections and high-speed transmission capabilities of 5G networks are the strong support for the construction of smart cities.However,under the condition of continuous improvement and upgrading of 5G signal base stations and 5G mobile equipment facilities,the utilization rate of 5G network of China Mobile equipment is still low,and it has not yet reached the large-scale connection conditions of 5G network required for the construction of smart cities in China.Based on this development status,this thesis uses the real mobile user dataset with a total of 140,000 samples from the mobile official,and uses the machine learning algorithm to identify the mobile user dataset.Modeling a large amount of user data through algorithms helps identify potential 5G users and conduct accurate marketing,helping to improve the utilization rate and large-scale connection of 5G networks,which is of great significance for China’s smart city construction and 5G network commercialization.This thesis first preprocesses the original data,including missing value processing,data binning,feature scaling,and data structure optimization.Then,the F1 score is used as the evaluation index for algorithm classification and recognition,and the principles of K-Nearest Neighbor,XGBoost,CatBoost,LightGBM,MLP and other algorithms are introduced,and the recognition effects of various algorithms on the data before and after SMOTE oversampling are tested.It is found that after the original mobile user imbalanced dataset is processed by SMOTE oversampling,the recognition performance of various algorithms for 5G users is significantly improved.The weighted F1 score of the optimized XGBoost algorithm is 0.8801,which is significantly better than the K-Nearest Neighbor and Random Forest algorithms.Finally,the XGBoost algorithm is optimized by grid search parameter tuning and stacking integration,and the final 5G potential user identification model is obtained,and its weighted F1 score is 0.8851.After analyzing the identification results,the following conclusions are drawn:(1)For the unbalanced dataset of mobile users,the SMOTE oversampling method can optimize its data structure and significantly improve the recognition performance of the algorithm.(2)In the problem of 5G potential user identification,the classification and identification performance of the XGBoost algorithm is better than that of Logistic Regression,K-Nearest Neighbor,SVM,Decision Trees,Random Forests,GBDT and other algorithms.(3)Grid search parameter adjustment and stacking integration optimization can greatly improve the recognition performance of XGBoost algorithm for 5G users.(4)The weighted F1 score of the Stacking integrated model based on the XGBoost optimization algorithm reaches 0.8851,the recognition F1 score of 5G users is 0.7233,and the recall rate is 0.7216.The recognition effect is relatively good,and it can be used as a recognition model for potential 5G users. |
