[Multiple dependent variables LS-SVM regression algorithm and its application in NIR spectral quantitative analysis].

Area of Science:

• Chemometrics
• Spectroscopy
• Machine Learning

Context:

• Near-Infrared (NIR) spectroscopy is widely used for quantitative analysis of chemical constituents.
• Simultaneous analysis of multiple components often requires sophisticated modeling techniques.
• Optimizing model parameters is crucial for accurate quantitative predictions.

Purpose:

• To develop and validate a Multiple Dependent Variables Least Squares Support Vector Machine (MLS-SVM) regression model.
• To investigate the impact of optimized weights on model performance for multi-component analysis.
• To compare the performance of MLS-SVM with the standard LS-SVM algorithm for NIR quantitative analysis.

Summary:

• A novel MLS-SVM algorithm was developed, allowing for optimized weights in regression models.
• The model was applied to simultaneously analyze protein, lysine, and starch content in broomcorn using NIR spectra.
• Leave-one-out cross-validation was employed to determine optimal weights and model parameters, minimizing average relative error.

Impact:

• MLS-SVM demonstrated comparable performance to LS-SVM in multi-component quantitative analysis, achieving low average relative errors (1.37-6.47%) and high correlation coefficients (0.8392-0.9940).
• The study validates the necessity of optimizing weights in MLS-SVM for improved prediction accuracy in multiple dependent variable NIR modeling.
• This research introduces a new chemometric approach for simultaneous multi-component quantitative analysis using NIR spectroscopy.