Document Type thesis Author Name Shulga, Yelena A URN etd-011106-150017 Title Model-based calibration of a non-invasive blood glucose monitor Degree MS Department Mathematical Sciences Advisors Jayson D. Wilbur, Advisor Keywords robust regression partial least squares regression glucose non-invasive ordinary least squared regression ridge regression local regression weighted least squares regression diabetes Date of Presentation/Defense 2005-12-12 Availability unrestricted
This project was dedicated to the problem of improving a non-invasive blood glucose monitor being developed by the VivaScan Corporation. The company has made some progress in the non-invasive blood glucose device development and approached WPI for a statistical assistance in the improvement of their model in order to predict the glucose level more accurately. The main goal of this project was to improve the ability of the non-invasive blood glucose monitor to predict the glucose values more precisely. The goal was achieved by finding and implementing the best regression model. The methods included ordinary least squared regression, partial least squares regression, robust regression method, weighted least squares regression, local regression, and ridge regression. VivaScan calibration data for seven patients were analyzed in this project. For each of these patients, the individual regression models were built and compared based on the two factors that evaluate the model prediction ability. It was determined that partial least squares and ridge regressions are two best methods among the others that were considered in this work. Using these two methods gave better glucose prediction. The additional problem of data reduction to minimize the data collection time was also considered in this work.
Browse by Author | Browse by Department | Search all available ETDs
Questions? Email firstname.lastname@example.org