Optimization of Parameter Selection for Partial Least Squares Model Development.

Area of Science:

• Chemometrics
• Spectroscopy
• Data Analysis

Background:

• Multivariate calibration models require optimization of nonsystematic parameters.
• Selecting spectral pretreatment, variable importance in projection (VIP), and latent factors is challenging.

Purpose of the Study:

• To develop a novel and systematic approach for optimizing multivariate calibration models.
• To utilize a processing trajectory for parameter selection in Partial Least-Square (PLS) models.

Main Methods:

• A processing trajectory was employed to select spectral pretreatments, VIP for variable selection, and latent factors.
• Simultaneous assessment of calibration and prediction errors (RMSEC, RMSEP), RPD, and coefficients (Rcal(2), Rpre(2)) was performed.
• Three distinct near-infrared (NIR) datasets were used for validation.

Main Results:

• The developed method identified multiple optimal modeling paths across different datasets.
• The robust Partial Least-Square (PLS) model demonstrated superior efficiency compared to existing methods.
• The processing trajectory approach systematically optimized key model parameters.

Conclusions:

• A systematic processing trajectory offers an efficient method for optimizing multivariate calibration models.
• This approach provides a robust alternative to step-by-step optimization, enhancing quantitative analysis.
• Multiple optimal modeling pathways can exist for spectral datasets.