A hybrid variable selection method based on uninformative variable elimination and discretized wild horse...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Spectral data analysis is crucial in modern analytical chemistry.
• High-dimensional spectral data presents challenges for traditional variable selection methods.
• Existing methods often have limitations in efficiency and accuracy for complex samples.

Purpose of the Study:

• To develop an efficient and accurate spectral variable selection strategy.
• To address the limitations of single variable selection methods in high-dimensional spectral data.
• To establish a robust prediction model using a reduced variable subset.

Main Methods:

• A hybrid strategy combining Uninformative Variable Elimination (UVE) and Discretized Wild Horse Optimizer (DWHO) was proposed.
• UVE was used to eliminate noise variables, followed by DWHO for further refinement.
• Partial Least Squares (PLS) models were constructed using the selected variable subsets (UVE-DWHO-PLS).

Main Results:

• The UVE-DWHO-PLS method reduced the number of selected variables by over 85% compared to full-spectrum analysis.
• Prediction accuracy remained high, with R values consistently above 0.95 across four diverse datasets (orange juice, diesel, wine, blood).
• Lower Root Mean Squared Error of Prediction (RMSEP) values were achieved, indicating improved model performance.

Conclusions:

• The UVE-DWHO hybrid strategy offers a valuable approach for spectral variable selection in complex samples.
• This method enables the establishment of stable and accurate prediction models with significantly fewer variables.
• It overcomes limitations of single variable selection methods in high-dimensional spectral analysis.