Error propagation of partial least squares for parameters optimization in NIR modeling.

Area of Science:

• Chemometrics
• Analytical Chemistry
• Spectroscopy

Background:

• Partial Least Squares (PLS) is widely used in Near-Infrared (NIR) modeling.
• Model accuracy heavily relies on parameter selection.
• Understanding error propagation is crucial for robust model development.

Purpose of the Study:

• To propose a novel methodology for determining error propagation in PLS modeling parameters.
• To evaluate the impact of spectral pretreatment, latent variables, and variable selection on PLS model errors.
• To provide guidance for optimizing PLS model parameters in NIR analysis.

Main Methods:

• Developed a methodology to quantify error propagation in PLS parameters.
• Applied the methodology to establish PLS models using corn and Gardenia datasets.
• Analyzed error propagation using Type I and Type II errors for water and geniposide quantification.
• Compared different variable selection algorithms (VIP, iPLS, BiPLS, SiPLS).

Main Results:

• Demonstrated the significant impact of modeling parameters on PLS error propagation.
• Quantified error weight variations (e.g., 5% to 65%) across different variable selection methods for geniposide.
• Established that higher error weight correlates with poorer model performance.
• Successfully developed robust PLS models for corn and Gardenia using optimal parameters.

Conclusions:

• The proposed methodology effectively reveals how modeling parameters influence PLS error propagation.
• Optimal parameter selection is critical for developing accurate and reliable NIR models.
• This work offers valuable insights for selecting parameters in PLS and other multivariate calibration models.