A hybrid variable selection method combining Fisher's linear discriminant combined population analysis and an...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Accurate composition measurement is crucial for industrial process control.
• Near-infrared (NIR) spectroscopy offers a rapid, non-destructive analytical technique.
• Variable selection is essential for building robust chemometric models, especially with multicollinear spectral data.

Purpose of the Study:

• To propose a novel hybrid variable selection method for near-infrared (NIR) spectroscopy-based model building.
• To enhance the accuracy and reliability of composition measurements in industrial processes.
• To address challenges posed by multicollinearity in spectral data.

Main Methods:

• A double-layer variable selection strategy combining Fisher's linear discriminant combined population analysis (FCPA) and an improved binary cuckoo search algorithm (IBCS).
• FCPA for initial rough localization of informative variable intervals, handling multicollinearity.
• IBCS, enhanced with opposition-based learning (OBL) and jumping genes (JG), for fine selection of key variables, avoiding local optima.

Main Results:

• The proposed FCPA-IBCS method demonstrated superior performance in variable selection compared to other methods.
• Partial Least Squares (PLS) models built using FCPA-IBCS showed higher fitting accuracy.
• Reduced prediction errors were observed for the calibration models predicting beer extract, corn protein/starch, and diesel fuel boiling point.

Conclusions:

• The hybrid FCPA-IBCS variable selection method is effective for NIR spectroscopic analysis in industrial applications.
• This approach improves the accuracy and robustness of chemometric models for composition measurement.
• The method successfully navigates multicollinearity and optimizes variable selection for enhanced predictive performance.