Machine Learning-Enhanced Evaluation of Handheld Laser-Induced Breakdown Spectroscopy (LIBS) Analytical Performance...

Area of Science:

• Geochemistry
• Analytical Chemistry
• Spectroscopy

Background:

• Handheld laser-induced breakdown spectroscopy (hLIBS) is an emerging technique for in situ rock analysis.
• hLIBS instruments provide qualitative and quantitative geochemical data, but performance varies between manufacturers.
• Effective calibration methods are crucial for reliable hLIBS data.

Purpose of the Study:

• To compare the analytical performance of two commercial hLIBS instruments.
• To evaluate the effectiveness of multivariate calibration methods (PLS, RF, ANN) against univariate analysis for geochemical characterization.
• To determine the impact of spectral resolution on hLIBS quantitative analysis.

Main Methods:

• Utilized two commercial hLIBS instruments with noise reduction and partial least squares (PLS) calibration.
• Applied Leave-One-Out Cross-Validation (LOOCV) for model validation.
• Explored Random Forest (RF) and Artificial Neural Network (ANN) algorithms to model nonlinear relationships, comparing them with univariate analysis.

Main Results:

• Multivariate methods significantly outperformed univariate analysis for geochemical quantification.
• Pearson's coefficient (R²) and root-mean-square error (RMSE) were used to assess predictive performance on 21 certified reference materials.
• Spectral resolution was identified as the primary factor influencing multivariate calibration performance.

Conclusions:

• Partial least squares (PLS) calibration was effective for the higher-spectral-resolution hLIBS instrument.
• Complementary algorithms (RF, ANN) were necessary to achieve optimal results with the lower-spectral-resolution instrument.
• hLIBS, particularly with high spectral resolution and appropriate multivariate calibration, is a powerful tool for in situ geochemical analysis.