Predicting App Launches on Mobile Devices Using Intelligent Agents and Machine Learning

Abstract

Data rich applications often have to load large amounts of data upon launch. The launch times for these applications, e.g. Facebook and NU.nl, can be improved by prefetching their data prior to use. This requires reliable predictions on what applications the user will use in the near future. In order to perform successful predictions, this research utilizes intelligent agents and reinforcement learning. With it, the devised system is able to successfully predict 45.6% of all applications launched by a user. The intelligent agent framework Jadex provides the communication between the agents and Q-learning is used along with the time-of-day as a reinforcement learning algorithm. The results are obtained via simulations with the LiveLab dataset which contains phone usage from 24 users over about a year time. The flexible MAS allows for many improvements in future work that promise even better results.