Comprehensive comparison on different wavelength selection methods using several near-infrared spectral datasets with...

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Near-Infrared (NIR) spectroscopy is vital in fields like agriculture and food safety.
• High dimensionality and data redundancy in NIR spectra present significant analytical challenges.
• Effective wavelength selection is crucial for optimizing NIR spectroscopic analysis.

Purpose of the Study:

• To evaluate and compare the performance of diverse wavelength selection methods for NIR spectral datasets.
• To provide a comprehensive reference for researchers selecting optimal wavelength selection techniques.
• To identify the most effective methods for handling high-dimensional NIR data.

Main Methods:

• Categorization of wavelength selection methods into four groups: Partial Least Squares (PLS) parameter-based, Intelligent Optimization Algorithms (IOA)-based, Model Population Analysis (MPA)-based, and Wavelength Interval Selection (WIS) methods.
• Comparative analysis based on metrics including R²_C, R²_P, RMSEC, RMSEP, selected variables, computational time, and iRMSEP.
• Evaluation of twenty characteristic wavelength selection methods, including bootstrapping soft shrinkage (BOSS) and genetic algorithm interval partial least squares (GA-iPLS).

Main Results:

• Models developed using MPA-based and WIS methods demonstrated superior stability and performance across most datasets compared to other categories.
• Among the twenty methods assessed, bootstrapping soft shrinkage (BOSS) and genetic algorithm interval partial least squares (GA-iPLS) exhibited the best overall performance.
• The study highlights the effectiveness of specific advanced methods in mitigating challenges associated with high-dimensional NIR data.

Conclusions:

• MPA-based and WIS wavelength selection methods are highly recommended for robust NIR spectral analysis.
• BOSS and GA-iPLS emerge as top-performing techniques for wavelength selection in complex NIR datasets.
• This research provides valuable insights for optimizing analytical strategies in NIR spectroscopy applications.