Predicting Replacement of Smartphones with Mobile App Usage

To identify right customers who intend to replace the smartphone can help to perform precision marketing and thus bring significant financial gains to cellphone retailers. In this paper, we provide a study of exploiting mobile app usage for predicting users who will change the phone in the future. We first analyze the characteristics of mobile log data and develop the temporal bag-of-apps model, which can transform the raw data to the app usage vectors. We then formularize the prediction problem, present the hazard based prediction model, and derive the inference procedure. Finally, we evaluate both data model and prediction model on real-world data. The experimental results show that the temporal usage data model can effectively capture the unique characteristics of mobile log data, and the hazard based prediction model is thus much more effective than traditional classification methods. Furthermore, the hazard model is explainable, that is, it can easily show how the replacement of smartphones relate to mobile app usage over time.