Context aware model-based cleaning of data streams

Despite advances in sensor technology, there are a number of problems that continue to require attention. Sensors fail due to low battery power, poor calibration, exposure to the elements and interference to name but a few factors. This can have a negative effect on data quality, which can however be improved by data cleaning. In particular, models can learn characteristics of data to detect and replace incorrect values. The research presented in this paper focuses on the building of models of environmental sensor data that can incorporate context awareness about the sampling locations. These models have been tested and validated both for static and streaming data. We show that contextual models demonstrate favourable outcomes when used to clean streaming data.