|
Variable subset selection in linear regression |
|
|
|
|
Written by Agnar Höskuldsson, DTU
|
Many companies are collecting large amount of data and frequently it is a basic issue, which part of data should be used in the analysis. For instance, companies are investing in NIR instruments for on-line control of the production. Which part of the spectra should be used in the analysis?
Many companies are basing their instruments on the NIR technology. Foss-Electric (here shortened to Foss) in Denmark is selling measurement instruments based on  the NIR technology for more than 200 mio euros a year. Each instrument is calibrated to measure what it is designed for. The instrument selects appropriate parts of the spectra and by using a linear regression analysis it estimates typically between three and seven chemical parameters. The company has successfully implemented the technology in many different types of instruments. An important question is: Is it possible to establish an automatic procedure that finds the appropriate parts of the spectra that should be used such that the regression analysis gives efficient results? This is the question that will be treated in the following Tutorial.
|
