A Comparative Investigation of the Combined Effects of Pre-Processing, Wavelength Selection, and Regression Methods...

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Near-infrared (NIR) spectroscopy is a versatile analytical technique used across industries like pharmaceuticals and food.
• Its advantages include non-destructive analysis, speed, and minimal sample preparation.
• Successful NIR calibration depends on spectral data pre-processing, wavelength selection, and regression modeling.

Purpose of the Study:

• To comparatively study the interactions among pre-processing, wavelength selection, and regression methods in NIR spectral calibration.
• To evaluate the role of each aspect in achieving optimal calibration models.
• To identify effective combinations of methods for superior calibration performance.

Main Methods:

• Four benchmark datasets were utilized.
• Three pre-processing methods: Orthogonal Signal Correction (OSC), Extended Multiplicative Signal Correction (EMSC), Optical Path-Length Estimation and Correction (OPLEC).
• Two wavelength selection methods: Stepwise Forward Selection (SFS), Genetic Algorithm optimization with Partial Least Squares (GAPLSSP).
• Four regression methods: Partial Least Squares (PLS), LASSO, Least Squares Support Vector Machine (LS-SVM), Gaussian Process Regression (GPR).

Main Results:

• Pre-processing of spectral data was found to play a significant role in calibration.
• Wavelength selection demonstrated a marginal impact on calibration performance.
• The combination of specific pre-processing, wavelength selection, and nonlinear regression methods yielded superior results compared to traditional linear methods.

Conclusions:

• Effective spectral data pre-processing is crucial for robust NIR calibration.
• Nonlinear regression methods, when combined with appropriate pre-processing and wavelength selection, outperform linear approaches.
• This study provides insights into optimizing NIR calibration strategies by understanding method interactions.