An efficient variable selection method based on random frog for the multivariate calibration of NIR spectra.

Area of Science:

• Chemometrics
• Spectroscopy
• Data Science

Background:

• Variable selection is crucial for building accurate spectrum models.
• Existing methods may not optimally identify informative spectral regions.
• Near-infrared (NIR) spectroscopy requires robust variable selection techniques.

Purpose of the Study:

• To develop a novel variable interval selection method called Interval Selection based on Random Frog (ISRF).
• To enhance the prediction performance and interpretability of spectrum models.
• To evaluate the efficiency of ISRF compared to other established methods.

Main Methods:

• Developed the Interval Selection based on Random Frog (ISRF) algorithm.
• Utilized the Random Frog (RF) algorithm for initial informative variable identification.
• Applied a local search strategy to expand informative variable intervals.
• Tested ISRF on three near-infrared (NIR) datasets.

Main Results:

• ISRF effectively identified the most informative spectral intervals.
• The proposed ISRF method demonstrated superior performance compared to GA-PLS, RF, iRF, and iVISSA.
• ISRF significantly improved the prediction accuracy of the NIR models.
• Enhanced model interpretability was observed using the ISRF method.

Conclusions:

• ISRF is a highly efficient and effective method for variable interval selection in spectrum modeling.
• The ISRF approach offers advantages in both predictive power and interpretability for NIR data.
• This novel method provides a valuable tool for chemometric analysis and spectral data modeling.