An effective quality measure for prediction of context information

Pervasive systems must be able to adapt themselves to changing environments to ensure the QoS for the user and to optimize resource usage. In this respect, acting proactively before an event has occurred, like starting to heat the house based on the prediction that the user will arrive within the next hour, can yield better results than reacting at the moment it occurs. Especially for context prediction, the quality of context is important since a wrong prediction can be costly and diminish user satisfaction. In this work we evaluate algorithms with respect to the quality of their inference for higher-level context information and describe how these are realized as plug-ins for a middleware that enables context-aware, self-adaptive applications.