Modeling with Multiple Correlated Spectral Data Based on Approximating the Nonlinear Spectrum Induced by Scattering.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Scattering in solutions introduces nonlinearity, hindering accurate spectral quantitative analysis.
• Existing methods struggle to model scattering characteristics effectively, limiting prediction accuracy.
• Nonlinearity affects modeling variables and data representation in spectral analysis.

Purpose of the Study:

• To develop a novel method for spectral quantitative analysis that mitigates nonlinearity caused by scattering.
• To enhance prediction accuracy in the analysis of scattering solutions.
• To improve the robustness of spectral models against scattering effects.

Main Methods:

• Combined spectral data from multiple, equally spaced optical pathlengths as a modeling dataset.
• Utilized Partial Least Squares (PLS) modeling with increased principal component selection options.
• Applied reduced weighting to wavelengths significantly affected by scattering to enhance model insensitivity.

Main Results:

• The proposed method demonstrated improved prediction accuracy compared to traditional and normalization techniques.
• Achieved a 61.7% increase in prediction accuracy over traditional methods.
• Showed a 58.5% improvement compared to variable sorting for normalization methods.

Conclusions:

• The developed method effectively addresses nonlinearity issues in spectral quantitative analysis of scattering solutions.
• Combining spectral data from varied pathlengths and weighted wavelength selection enhances model accuracy and reliability.
• The approach is feasible and offers significant improvements for analyzing strongly scattering materials.