Comparison between Variable-Selection Algorithms in PLS Regression with Near-Infrared Spectroscopy to Predict...

Area of Science:

• Environmental Science
• Analytical Chemistry
• Geochemistry

Background:

• Soil metal contamination poses environmental risks.
• Traditional soil metal analysis is costly and time-consuming.
• Near-infrared (NIR) spectroscopy offers a potential alternative for rapid analysis.

Purpose of the Study:

• To evaluate the performance of the Firefly algorithm by intervals in partial least-squares (FFiPLS) for predicting soil metal content.
• To compare FFiPLS against deterministic partial least-squares (PLS) algorithms.
• To assess the effectiveness of NIR spectroscopy coupled with chemometrics for soil metal analysis.

Main Methods:

• Collected NIR spectra (1000-2500 nm) from river basin soil samples.
• Developed predictive chemometric models using PLS, interval-PLS (iPLS), successive projections algorithm for interval selection in PLS (iSPA-PLS), and FFiPLS.
• Applied data preprocessing techniques including multiplicative scatter correction (MSC) and standard normal variate (SNV).

Main Results:

• FFiPLS models achieved a relative prediction deviation (RPD) > 2 for iron and titanium.
• Adequate models (RPD > 2) were developed for aluminum using various preprocessing methods and raw data.
• FFiPLS outperformed deterministic algorithms for predicting aluminum, beryllium, gadolinium, and yttrium content.

Conclusions:

• FFiPLS is a robust variable selection technique for soil metal prediction using NIR spectroscopy.
• Chemometric models, particularly FFiPLS, can provide accurate and efficient determination of metals in soil.
• This approach offers a viable alternative for environmental monitoring of soil metal contamination.