In many chemometric and biological applications, responses of interest are expensive to measure (eg ripeness of fruit or nitrogen content of manure), whereas explanatory variables are cheap and quick to measure, such as those arising from spectroscopy.
These variables are often highly dependent and numerous, as many as several hundreds. Motivated by a problem of predicting the ripeness of fruit from infra-red spectroscopic data from the UK and Australia, a variety of statistical methods have been investigated and compared, including least-squares support vector machines and partial least squares.
The project is led by Alan Kimber(S3RI) with PhD student Carmen Ybarra Moncada and data from biologists from the UK and Australia.