Effective Quality-Aware Sensor Data Management

Abstract

Sensor networks are widely used for heterogeneous applications in diverse domains. One sensor network may provide data for applications and services that often have different or even competing requirements on the quality of the gathered sensor data; for example, the level of precision requested, the time constraints for getting the data, the level of accuracy, and so on. Providing the required level of quality for all applications and users is difficult, hindered by potential sensor malfunctions, communication problems, tampering, environmental conditions, and so on. Here, we propose a quality-aware sensor data management framework, which allows different users to define their own quality requirements by using a “virtual” sensor. These virtual sensors attempt to provide users with sensor data that satisfies their requests. For missing sensor readings or low quality data, association rule mining is used to estimate missing values, thus, improving users’ perceived quality. Experiments conducted on a real sensor dataset show promising results in the estimation of missing sensor readings, compared to other data imputation techniques, when a significant spatio-temporal correlation among sensors exists.