Rapid technological innovation has accelerated the diffusion of mobile phones and communication networks, making phone replacement an important consumer behavior in the telecommunications market. However, limited research has examined the determinants of phone replacement decisions using large-scale behavioral data from telecom operators. This study aims to identify the key predictive factors influencing users' phone replacement behavior and develop a predictive framework for telecom operators.
Using panel data from approximately 400,000 telecom users, this study develops a prediction framework based on the XGBoost machine learning algorithm. The model incorporates large-scale behavioral data, including users' demographic characteristics, phone attributes and mobile communication behaviors. To enhance model interpretability, SHAP analysis is employed to assess the importance and interaction effects of predictive features.
The results show that historical phone replacement behavior is the strongest predictor of future replacement decisions. Brand prominence of the currently owned phone is negatively associated with replacement likelihood, suggesting that users with well-known brands tend to retain their devices longer. In addition, communication demand indicators, such as call duration and SMS usage, positively influence phone replacement probability, with heterogeneous effects across different user groups.
This study integrates large-scale telecom behavioral data with explainable machine learning techniques to predict consumer phone replacement behavior, providing a transparent analytical framework and practical insights for telecom operators to design personalized marketing strategies.
