Characteristic wavelength optimization for partial least squares regression using improved flower pollination...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Near-infrared (NIR) spectroscopy is vital for multivariate analysis.
• Effective wavelength selection is key to improving model generalization and reducing complexity.
• Existing methods may not fully optimize spectral variable selection for complex datasets.

Purpose of the Study:

• To introduce a novel wavelength selection technique, the interval flower pollination algorithm (iFPA).
• To apply iFPA for spectral variable selection within Partial Least Squares Regression (PLSR) models.
• To evaluate the performance and interpretability of iFPA on diverse NIR datasets.

Main Methods:

• Development of the interval flower pollination algorithm (iFPA) with three distinct phases.
• Phase 1: Flower Pollination Algorithm for informative spectral variable identification.
• Phase 2: Variable elimination and local search for optimal continuous spectral intervals.
• Phase 3: Application and validation on corn, diesel, and soil NIR datasets using PLSR.

Main Results:

• The iFPA demonstrated superior prediction performance compared to other wavelength selection methods.
• Achieved low root mean square error of prediction (RMSEP) values across different datasets (corn: 0.0096-0.0727, diesel: 0.0015-3.9717, soil: 1.3388-29.1144).
• Selected variables showed good interpretability, aiding in understanding the spectral analysis.

Conclusions:

• The proposed iFPA is an effective method for spectral variable selection in NIR spectroscopy.
• iFPA combined with PLSR offers enhanced prediction accuracy and model interpretability.
• This approach holds significant potential for advancing quantitative analysis in various fields using NIR data.