Managing uncertainty in sensor databases

Abstract

Sensors are often employed to monitor continuously changing entities like locations of moving objects and temperature. The sensor readings are reported to a centralized database system, and are subsequently used to answer queries. Due to continuous changes in these values and limited resources (e.g., network bandwidth and battery power), the database may not be able to keep track of the actual values of the entities, and use the old values instead. Queries that use these old values may produce incorrect answers. However, if the degree of uncertainty between the actual data value and the database value is limited, one can place more confidence in the answers to the queries. In this paper, we present a framework that represents uncertainty of sensor data. Depending on the amount of uncertainty information given to the application, different levels of imprecision are presented in a query answer. We examine the situations when answer imprecision can be represented qualitatively and quantitatively. We propose a new kind of probabilistic queries called Probabilistic Threshold Query, which requires answers to have probabilities larger than a certain threshold value. We also study techniques for evaluating queries under different details of uncertainty, and investigate the tradeoff between data uncertainty, answer accuracy and computation costs.