[A new wavelength selection algorithm based on the fusion of multiple models].

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Context:

• Near-Infrared (NIR) spectroscopy generates large datasets with numerous wavelengths.
• Many wavelengths in NIR spectra lack informative value or contain noise, negatively impacting model complexity and predictive accuracy.
• Effective variable selection is crucial for optimizing NIR models.

Purpose:

• To address the limitations of single Partial Least Squares (PLS) regression models for variable selection in NIR spectroscopy.
• To propose and validate a novel method for wavelength selection based on the fusion of multiple PLS models.
• To improve the accuracy and stability of variable selection in chemometric modeling.

Summary:

• The paper reviews existing variable selection methods using single PLS models, highlighting their inaccuracies and instability.
• A new method is proposed that fuses regression coefficients from multiple PLS models to identify and eliminate non-informative wavelengths.
• This fused approach utilizes a threshold to filter variables, leading to more robust model development.

Impact:

• The proposed method demonstrates potential for creating NIR models with reduced complexity and enhanced predictive capabilities.
• Experimental validation on three NIR datasets shows competitive or superior performance compared to UVE-PLS and GA-PLS algorithms.
• The method is efficient, elegant, and suitable for practical application in NIR data analysis.