With the popularity of smart devices, application responsiveness is one of the most important indicators of user experience, and forecast the next application will be used by users is crucial to the management planning for related companies. In this paper, more sophisticated machine learning, and even deep learning, which can achieve better forecast performance, is widely used in this field. However, the opacity of these black-box models greatly limits user trust and how well developers can optimize the model. To address these issues, this paper first tests six of the most popular forecasting algorithms, including ensemble models and neural networks, to select the optimal model. As an innovation, this paper also uses XAI techniques to explain black-box models to increase user trust in the results generated by forecast models and to help developers in their work. After completing the above work, on the one hand, we found that the ensemble model performed better in the time series datasets with user application usage information, especially with LightGBM, on the other hand, we found that the prediction model using the SHAP method showed that the target variable categorical features and lags Features are important features to forecast the user's next application.