Residual compensation algorithm for machine learning-based LIBS quantification optimization.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Materials Science

Background:

• Quantitative analysis using Laser-Induced Breakdown Spectroscopy (LIBS) often faces challenges with prediction accuracy.
• Environmental and sample-specific parameters can significantly influence LIBS spectral data and subsequent quantitative predictions.
• Existing models may not fully account for these variables, leading to prediction errors.

Purpose of the Study:

• To develop and validate a novel quantitative prediction algorithm for LIBS based on residual compensation.
• To enhance prediction accuracy by integrating environmental and sample parameters into the LIBS quantitative model.
• To evaluate the effectiveness of the proposed algorithm across different regression models and sample types.

Main Methods:

• A residual compensation algorithm was developed for LIBS quantitative prediction.
• The algorithm was integrated with Support Vector Machine Regression (SVR), Partial Least Squares Regression (PLSR), Random Forest Regression (RFR), and K-Nearest Neighbor Regression (KNNR) models.
• A 10-fold cross-validation approach was employed using aluminum alloy samples with 10 elements.

Main Results:

• The residual compensation algorithm significantly reduced Mean Absolute Error of Prediction (MAEP) and Mean Relative Error of Prediction (MREP).
• Compared to the original PLSR model, MAEP and MREP were reduced by an average of 51.8% and 64.8%, respectively.
• For the SVR-based model, MAEP and MREP were reduced by 43.0% and 51.1%, respectively, demonstrating broad applicability.

Conclusions:

• The proposed residual compensation algorithm effectively enhances LIBS quantitative prediction accuracy.
• Incorporating environmental and sample parameters is crucial for improving LIBS analytical performance.
• This method offers a robust approach for more reliable elemental analysis in aluminum alloys using LIBS.