Abstract
This thesis presents the Life Cycle Prediction Model (LCPM), a probabilistic model that models the time when a context of a smartphone will remain the same as a life cycle. A correct prediction allows the corresponding sensors on the smartphone to be turned off during the life time of the context. Thus, it is possible to eliminate periodic and redundant sampling of the sensors, resulting in energy saving. Different ways of building the probabilistic LCPM for a given context are discussed, which try to trade off between miss predictions and energy consumption. We also address the issues resulting from miss predictions by identifying possible causes and proposing feasible solutions. Our experiments using data collected from real users on various contexts show that LCPM can adapts to different kinds of contexts and results in significant power saving.