A Two-Step Simulated Annealing Algorithm for Spectral Data Feature Extraction.

Area of Science:

• Spectral analysis
• Machine learning
• Chemometrics

Background:

• Traditional spectral feature extraction methods often suffer from low modeling accuracy and poor stability.
• Addressing these limitations is crucial for reliable data analysis in various scientific applications.

Purpose of the Study:

• To develop an improved spectral feature extraction algorithm that enhances model accuracy and stability.
• To introduce a novel approach combining global and local optimization techniques for feature selection.

Main Methods:

• The proposed two-step simulated annealing algorithm (TSSA) integrates the Boruta algorithm for feature selection with simulated annealing for optimization.
• The Boruta algorithm identifies features strongly correlated with the dependent variable, minimizing data redundancy.
• The TSSA algorithm combines global and local optimization strategies for robust feature extraction.

Main Results:

• The TSSA algorithm demonstrated significantly improved accuracy and stability compared to traditional methods.
• The inversion model built using TSSA achieved a high coefficient of determination (R² = 0.9654).
• Excellent performance was indicated by a root mean square error (RMSE) of 3.6723 μg/L and a mean absolute error (MAE) of 3.1461 μg/L.

Conclusions:

• The TSSA algorithm offers a superior approach to spectral feature extraction, overcoming limitations of traditional methods.
• The enhanced accuracy and stability of TSSA-based models are validated by strong performance metrics.
• This method provides a reliable tool for building robust inversion models in spectral data analysis.