Metrohm Vision – Tutorial - Manual (Page 34)

1.7 Partial Least Squares Regression 
NIRS Vision Tutorial – Operation 
27
1 Loading
weights and coefficients 2 Statistical results
3 Statistical results
with statistical summary 4 PRESS and SEC
The program automatically selects the number of factors for the method based
on the approach of the PRESS (Predicted Residual Error Sum of Squares) value
to a minimum. The number of factors actually appropriate for a method should
be similar to the number of degrees of Freedom (chemical and physical
variables) in the sample set. Thus, varying temperature and constituent
concentrations represent variables that must be considered when selecting the
number of factors.
Remember, robustness of a model is often reduced as more factors are added.
The software selects the number of factors for the model to be the number of
factors which the PRESS value is the lowest for the first time. At this point, the
Standard Error of Cross Validation (SECV) should be visually compared with the
laboratory error for the Reference data used in developing the model. The
SECV should approach but not be lower than the lab error.
Save Equation
1. Click on Regression  Save Equation
2. Click on [OK]