Sorting the wheat from the chaff: Adaptive sensor selection for context-aware applications

Context-aware applications base their services on contextual information that can be queried from the sensors embedded in the environment. However, when the number of sensors and the applications using them increases, sensor query becomes on one hand a resource hungry task, e.g. for network bandwidth and energy needed to power the sensors, and on the other hand may yield loads of unnecessary information that should be processed by the context-aware application. Therefore, an informed sensor selection in such environments becomes a necessity. This paper proposes an algorithm for a relevance-based sensor query, which adaptively spends the allotted query budget on querying sensors that are most relevant to the user's concept. This relevance is measured using an objective function which combines both expected query cost and expected sensor utility, as observed from the the sensor query history. The results of our evaluation show the potential of our approach to approximate the user's concept with best accuracy while preserving the query budget.