Stacking and ridge regression-based spectral ensemble preprocessing method and its application in near-infrared...

Area of Science:

• Chemometrics
• Machine Learning
• Spectroscopy

Background:

• Spectral preprocessing is crucial for removing noise and enhancing prediction model performance.
• Existing methods often overlook complementary information from diverse preprocessing techniques, limiting model accuracy.
• Current ensemble methods may not fully exploit the potential of spectral data.

Purpose of the Study:

• To propose a novel spectral ensemble preprocessing method, stacking preprocessing ridge regression (SPRR), to address limitations of existing techniques.
• To leverage ensemble learning and ridge regression for improved spectral data utilization.
• To enhance the accuracy and reliability of prediction models using spectral data.

Main Methods:

• Applied multiple distinct spectral preprocessing techniques to original spectral data, creating diverse datasets.
• Trained individual ridge regression (RR) base models on each preprocessed dataset.
• Utilized RR as a meta-model to integrate base model outputs via stacking.

Main Results:

• Correlation analysis confirmed significant complementary information among differently preprocessed spectral data.
• SPRR demonstrated superior accuracy and reliability across six diverse datasets (apple, meat, wheat, olive oil, tablet, corn) compared to single methods and averaging ensembles.
• SPRR outperformed four common ensemble preprocessing methods under identical experimental conditions.

Conclusions:

• The proposed SPRR method effectively captures complementary information from various spectral preprocessing techniques.
• SPRR offers a robust and accurate approach for spectral data analysis, outperforming conventional and existing ensemble methods.
• This stacking ensemble approach provides a powerful tool for advancing chemometric and machine learning applications in spectroscopy.