Hierarchical extreme learning machine for dimensionality reduction in near-infrared spectral analysis

Area of Science:

• Spectroscopy
• Chemometrics
• Machine Learning

Background:

• Dimensionality reduction (DR) is essential for analyzing high-dimensional near-infrared (NIR) spectral data.
• Traditional DR methods face challenges in feature selection and model accuracy improvement.
• Effective DR is critical for robust chemometric modeling.

Purpose of the Study:

• To introduce a novel Hierarchical Extreme Learning Machine (HELM) for spectral DR.
• To enhance feature learning efficiency by reducing redundancy and improving feature independence.
• To develop an integrated HELM-Partial Least Squares (HELM-PLS) model for improved spectral data analysis.

Main Methods:

• Development of HELM utilizing mutually orthogonal weight vectors.
• Integration of HELM with Partial Least Squares (PLS) to form the HELM-PLS model.
• Validation of the HELM-PLS model on three public spectral datasets.

Main Results:

• The HELM-PLS model demonstrated excellent performance metrics (R, RMSE, RPD) across datasets.
• HELM-PLS exhibited superior prediction accuracy, robustness, and generalization compared to other DR methods.
• The HELM approach proved effective for spectral DR and feature extraction.

Conclusions:

• The HELM method offers a superior approach to spectral data DR and feature extraction.
• The HELM-PLS model provides an efficient and accurate predictive tool for complex spectral data analysis.
• This study highlights the potential of HELM in advancing chemometric applications.