| Author | Message | ||
     Vilas Vyankatrao Jangale (vilasmechgmailcom) Junior Member Username: vilasmechgmailcom Post Number: 10 Registered: 5-2010 |
Hello Venky, Sorry, I am not able to attach the file in pdf, as its size (800 kb) exceeds the size allowed (100 kb) for attachments in this forum. I have sent it to your email [email protected]. Please let me know if you would like me to send it to some other email address. Thank you, Vilas V Jangale | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 106 Registered: 3-2004 |
Can any body help me getting the paper " An example of 2-block predictive partial least squares regression with simulated data. Anal. Chim. Acta 185, 19-32 (1986)in pdf format " | ||
| Kelly Anderson |
I've been building a calibration model to predict L/S ratios, a fetal lung maturity test. From the residual plot based on a PLS model, there are positive deviations for low L/S values and negative deviations for higher L/S values. This suggests to me that there is some sort of systematic error involved in my model--most likely being in the spectra. Is there a preprocessing algorithm available that allows for the correction of systematic error? The data set has already been baseline corrected and mean-centered. | ||
| Chris Brown |
Kelly, By 'residual plot' do you mean a plot of the spectral model residuals, or the L/S prediction/fit residuals? If they are L/S residuals that you are speaking of, it may arise from any number of reasons. Briefly, one reason may simply be a crummy model (with useless model coefficients, your model will tend to predict the mean of the data, which can end up looking like over-predicting the low values and under-predicting the high values). Or, you may be suffering from 'slope-depression' (*sigh*), which is a result of having a large amount of unmodelable error carried into your model coefficients (regression vector). This is a so-called "errors-in-variables" problem. It may also just be reference error (although often this can be ruled out based on good analytical knowledge). All of these will result in a model which tends to over-predict low values and under-predict high values, but there are a number of other reasons as well. You might start by building your model on a subset of your calibration data, and seeing if the effect is even more pronounced. If it's spectral residuals, you've most likely got another matter on your hands, but I'm presuming it's L/S residuals you're speaking of. ~ C. | ||
| Richard Kramer |
It may be that you have systematic non-linearities which are important for a model. PLS will reject those very efficiently. Try PCR instead. If there are important, systematic non-linearities PCR may successfully exploit them to produce a better model than is possible with PLS for your particular data. |