Novel Variable Selection Method Based on Uninformative Variable Elimination and Ridge Extreme Learning Machine: CO...

Area of Science:

• Spectroscopy
• Chemometrics
• Data Analysis

Background:

• Variable selection is crucial in spectroscopy for accurate analysis.
• Traditional methods face challenges with interval selection and collinearity.
• Developing robust methods is essential for reliable spectral data interpretation.

Purpose of the Study:

• To propose a novel variable selection and assessment method for spectroscopy.
• To address limitations of traditional interval selection techniques.
• To enhance the accuracy and interpretability of spectroscopic analysis.

Main Methods:

• Utilized Uninformative Variable Elimination (UVE) to identify and remove irrelevant wavelengths.
• Employed Ridge Extreme Learning Machine (RELM) to handle collinearity and build predictive models.
• Applied Feature Selection Path (FSP) plot and Sparsity-Error Trade-off (SET) curve for optimal wavelength selection.

Main Results:

• UVE successfully identified informative wavelengths crucial for CO gas transmittance spectra.
• RELM demonstrated rapid modeling, effective collinearity management, and high accuracy (r=0.995 for CO retrieval).
• FSP and SET curves provided intuitive visualization for selecting optimal wavelength combinations.

Conclusions:

• The proposed UVE-RELM method offers an effective approach for variable selection in spectroscopy.
• RELM provides a robust and accurate alternative to traditional modeling methods.
• The visualization tools aid experts in wavelength selection and domain knowledge extraction.