By knowing which upcoming actions a user might perform, a mobile application can optimize a user interface or reduce the amount of user input needed for accomplishing a task. A herein-described prediction module can answer queries from a mobile application regarding which actions in the application the user is likely to perform at a given time. Any application can register and communicate with the prediction module via a straightforward application programming interface (API). The prediction module continuously learns a prediction model for each application based on the application's evolving event history. The prediction module generates predictions by combining multiple predictors with an online learning method, and capturing event patterns not only within but also across registered applications. The prediction module is evaluated using events collected from multiple types of mobile devices.