A retrospective look at cross model validation and its applicability in vibrational spectroscopy

Area of Science:

• Chemometrics
• Spectroscopic Analysis
• Analytical Chemistry

Background:

• Variable selection is crucial for building robust spectroscopic models.
• Traditional methods can be computationally intensive and may not ensure optimal predictive performance.
• Cross Model Validation (CMV) offers an efficient alternative for variable selection.

Purpose of the Study:

• To demonstrate the efficient application of Cross Model Validation (CMV) for variable selection in spectroscopic data.
• To validate the identified spectral regions against existing chemical literature.
• To showcase the stability and interpretability of models developed using CMV.

Main Methods:

• Application of Cross Model Validation (CMV), also known as double cross validation.
• Spectral pre-treatment: Standard Normal Variate (SNV) for FT-IR and 2nd derivative for NIR.
• Variable selection using jack-knifing and frequency of significance within CMV.

Main Results:

• High correspondence between CMV-identified wavelength bands and literature-reported chemical interpretations.
• Demonstrated model stability and consistent predictive performance through conservative validation.
• Successful use of down-weighting variables for enhanced prediction ability and model interpretability.

Conclusions:

• CMV is an effective technique for variable selection in spectroscopic applications, including FT-IR and NIR.
• The method objectively identifies relevant spectral regions, leading to interpretable and stable predictive models.
• CMV facilitates optimized model performance and detailed interpretation by down-weighting non-contributory variables.