With the accelerated digitization process, it is crucial to accurately identify potential 5G demand users through models to facilitate the transition from the 4G era to the 5G era in order to realize the construction of smart cities based on deep 5G applications. In order to accurately and efficiently identify 5G telecom users and analyze the business features that have important impacts on 5G users, this paper designs a 5G user prediction model combining k-means and classifier based on data mining. This model is based on k-means clustering to deal with the current dilemma of unbalanced sample size of 5G users and non-5G users, and additionally uses a bagging strategy to integrate two base classifiers, LightGBM and CatBoost, to effectively improve the model generalization capability. In comparison with the baseline model, the accuracy of the model is 87.7% and the F1-score is 70.4%. The results are satisfactory. With this model, 5G users can be identified with higher probability, which is a great help for mobile operators to identify potential users for accurate marketing. At the same time, the business characteristics highly correlated with 5G users also provide guidelines for 6G service design and package designation.