Author |
Message |
Eduardo Rocha
| Posted on Monday, June 30, 2003 - 1:02 pm: | |
I tried do make a PLS model for quantify additive in diesel by NIR spectroscopy. I knew it was a problem, because the additive was very low concentration (~200 ppm). The first results (24 samples/cross-validation) were not too bad: R=.988 and RMSEP=22ppm. However the loadings for first PC (88%)is too curious: almost all loadings are close to zero (in 445 variables, only 19 are greater than 0.05). If I follow the little I learning, these variables maybe deleted. But the model made with only 19 variables is worse than first one. I'd like to understand about the problem before the use in a test set. Anyone have an opinion ? Eduardo Rocha |
|