PCR vs. PLS (Part 8)

Trying to understand better the data set of meat meal, I gave the blue color to the samples from 0 to 10%, green to samples which has from 10% to 20%, orange color to samples from 20 to 30% and red to samples from 30 to 40%.

If we develop the PCA and apply the Mahalanobis Distance to different planes we get:

We can continue with the rest of combinations, trying to see which one gives the best  correlation or in this case the better pattern with the colors.

Developing the PCR Regression, we get the explained variance plot which can help us to see which PC has the highest correlation with the Ash constituent:

It is clear than 6th PC has the highest explanation for Ash, so let´s see the plane:

 

Ash is not an easy parameter to calibrate, and a high number of components are necessary for a calibration, but this one has the highest contribution in a PC regression in this case. Here are outliers that are not remove yet, but as soon they be removed, the correlation will improve a little bit.

 

> cor(NIR_princomp$scores[1:1280,6],Ash[1:1280,])
[1] 0.5487329

 

Really it is a high correlation for this parameter.