Bayesian Reasoning for Sensor Group-Queries and Diagnosis

As large-scale sensor networks are being deployed with the objective of collecting quality data to support user queries and decision-making, the role of a scalable query model becomes increasingly critical. An effective query model should scale well with large network deployments and address user queries at specified confidence while maximizing sensor resource conservation. In this paper, we propose a group-query processing scheme using Bayesian Networks (BNs). When multiple sensors are queried, the queries can be processed collectively as a single group-query that exploits inter-attribute dependencies for deriving cost-effective query plans. We show that by taking advantage of the Markov-blanket property of BNs, we can generate resource-conserving group-query plans, and also address a new class of diagnostic queries. Through empirical studies on synthetic and real-world datasets, we show the effectiveness of our scheme over existing correlation-based models.