DataSoap Project 

In traditional databases, either the data is input by an explicit data entry operation or the data is generated due to a transactional activity. However, In sensor databases, all data is gathered directly from sensors. This Live data from sensors needs to be cleaned for queries to make "sense." The DataSoap project is investigating techniques for on-line data cleaning, so that any actions taken based on this data is in fact correct and can be relied upon with high confidence.

Some of the techniques we are exploring include

1. Dealing with erroneous data
2. Recovering from missing data values
3. Outlier detection , error models, error measures
4. Calibration.
5. Gathering additional data to increase confidence

Papers

1. Book Chapter : Statistical approaches to cleaning sensor data, In a book on sensor networks, CRC press, November 2003
2. Elnahrawy and Badri Nath, Context aware sensors, To appear in European workshop on sensor networks, January 2004
3. Eiman Elnahrawy and Badri Nath, Cleaning and querying noisy sensors, In WSNA 2003, San Diego,