[Near infrared spectroscopy synergy interval wavelength selection method using the LSSVM model]

Area of Science:

• Spectroscopy
• Chemometrics
• Machine Learning

Context:

• Near-infrared (NIR) spectroscopy is crucial for analyzing complex samples like food products.
• Traditional wavelength selection methods often fail to account for nonlinear relationships within spectral data.
• Developing advanced algorithms is essential for maximizing the information extracted from NIR spectra.

Purpose:

• To introduce a novel wavelength selection algorithm, Synergy interval least squares support vector machines (siLSSVM).
• To address the limitations of existing methods by incorporating nonlinear factors.
• To improve the accuracy and robustness of chemometric models for spectral data analysis.

Summary:

• The proposed siLSSVM algorithm integrates interval strategies with synergy intervals, overcoming the limitations of traditional methods that ignore nonlinear factors.
• Applied to near-infrared spectral data of apple sugar, siLSSVM demonstrated significant improvements over standard Partial Least Squares (PLS) and Least Squares Support Vector Machines (LSSVM) models.
• Performance metrics showed substantial reductions in root-mean-square error of prediction (RMSEP) and increases in the correlative coefficient (RP).

Impact:

• siLSSVM efficiently identifies optimal wavelength intervals in spectral data exhibiting strong nonlinear characteristics.
• The algorithm significantly enhances prediction accuracy and model robustness for NIR spectral analysis.
• This provides a promising new approach for wavelength selection in complex spectral datasets, particularly those with nonlinear factors.