Boosting the Performance of Genetic Algorithms for Variable Selection in Partial Least Squares Spectral Calibrations

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Near-infrared (NIR) spectroscopy is a powerful tool for chemical analysis.
• Variable selection is crucial for building robust partial least squares (PLS) regression models.
• Genetic algorithms (GAs) are effective for optimizing complex models.

Purpose of the Study:

• To develop and evaluate a genetic algorithm (GA) incorporating adaptive boosting for variable selection in NIR spectroscopy.
• To identify informative wavelengths for improved PLS regression models.
• To demonstrate the advantages of the proposed method across diverse NIR spectral datasets.

Main Methods:

• A genetic algorithm (GA) was developed for variable selection in PLS regression.
• Adaptive boosting was integrated into the GA to enhance wavelength identification.
• The root mean square error of calibration (RMEC) was used as the fitness function.
• The performance was evaluated on three distinct NIR spectral datasets.

Main Results:

• The GA with adaptive boosting consistently outperformed the GA without adaptive boosting across all three datasets.
• Variable selected PLS models developed using the enhanced GA showed improved performance.
• The selected wavelengths by the GA with adaptive boosting captured relevant chemical information indicative of the analytes.

Conclusions:

• The integration of adaptive boosting into a GA significantly enhances wavelength selection for NIR-based PLS regression.
• This approach leads to more accurate and reliable quantitative and qualitative analysis.
• The developed method offers a robust solution for complex spectral data analysis.