A new spectral variable selection pattern using competitive adaptive reweighted sampling combined with successive...

Area of Science:

• Analytical Chemistry
• Chemometrics

Background:

• Multivariate calibration requires effective variable selection for accurate analysis.
• Existing methods like CARS and SPA have limitations in identifying optimal variable subsets.

Purpose of the Study:

• To introduce and evaluate the Competitive Adaptive Reweighted Sampling-Successive Projections Algorithm (CARS-SPA) for variable selection.
• To compare CARS-SPA performance against other established methods.

Main Methods:

• CARS-SPA combines Competitive Adaptive Reweighted Sampling (CARS) for initial selection and Successive Projections Algorithm (SPA) for refinement.
• Applied to Near-Infrared (NIR) data for nicotine and pesticide active ingredient analysis.

Main Results:

• CARS-SPA selected fewer, more informative variables compared to direct CARS.
• A Multiple Linear Regression (MLR) model using CARS-SPA variables showed superior prediction accuracy over PLS, SPA, and UVE-SPA models.
• Selected variable subsets captured essential chemical information, excluding redundant data.

Conclusions:

• CARS-SPA is a robust and efficient variable selection strategy for multivariate calibration.
• The method enhances predictive model performance by focusing on chemically relevant spectral information.