A Bayesian tree learning method for low-power context-aware system in smartphone

Context-aware services using smartphone have been proliferated for ubiquitous computing. However, the capacity of smartphone battery is extremely limited so that the services cannot be effectively used. In this paper, we propose a lowpower context-aware system using tree-structured Bayesian network. Bayesian network, one of the probabilistic models, is known to handle the uncertainty flexibly. A well-known problem of the probabilistic model, however, is high time complexity, which leads to significant consumption. To reduce the time complexity, we propose a tree-structure learning method. The key idea lies in how to consider the relation of each node. For the reason, we conduct the spanning tree based on the mutual information among nodes. The data for experiment were collected from Android phone for two weeks. The amount of the collected data is 7,464. The accuracy of proposed method achieves 94.13%. The energy consumption is measured using the power tutor application.