A New Local Modelling Approach Based on Predicted Errors for Near-Infrared Spectral Analysis

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Near-infrared spectroscopy (NIRS) and chemometrics are established analytical tools across industries.
• Many chemical processes exhibit complex multivariate and nonlinear characteristics, challenging traditional modeling.
• Accurate calibration models are crucial for reliable quantitative analysis in these complex systems.

Purpose of the Study:

• To develop an improved chemometric method for building accurate NIRS calibration models.
• To address the limitations of existing methods in handling nonlinear and multivariate data.
• To enhance both the predictive performance and computational efficiency of NIRS analysis.

Main Methods:

• A novel local errors regression method was developed.
• A new similarity criterion was introduced to select calibration subsets, utilizing spectral, chemical, and error information.
• Partial Least Squares (PLS) regression was applied to the selected subsets for model building.

Main Results:

• The proposed local errors regression method demonstrated superior performance compared to other local strategies.
• Significant improvements were observed in prediction ability for NIRS datasets.
• Enhanced calculation speed was achieved, making the method more efficient.

Conclusions:

• The local errors regression method offers a robust solution for developing accurate NIRS calibration models for complex analytes.
• This approach effectively handles multivariate and nonlinear data, improving analytical outcomes.
• The method provides a valuable advancement for NIRS applications in industries like pharmaceuticals.