Calibration transfer via an extreme learning machine auto-encoder.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Near-infrared (NIR) spectroscopy is widely used for chemical analysis.
• Spectra standardization is crucial for accurate quantitative analysis across different instruments and conditions.
• Existing methods like PDS, GLS, and CCA have limitations in handling complex spectral variations.

Purpose of the Study:

• To propose and evaluate a novel Transfer via Extreme learning machine Auto-encoder Method (TEAM) for NIR spectra standardization.
• To compare the performance of TEAM against established standardization techniques.
• To assess the efficacy of TEAM using diverse spectral datasets.

Main Methods:

• Development of the TEAM algorithm, integrating Extreme Learning Machine with Auto-encoder.
• Comparative analysis of TEAM with Piecewise Direct Standardization (PDS), Generalized Least Squares (GLS), and Canonical Correlation Analysis (CCA).
• Validation using three distinct spectral datasets: corn, tobacco, and pharmaceutical tablets.

Main Results:

• TEAM demonstrated superior stability and significantly reduced prediction errors compared to PDS, GLS, and CCA.
• TEAM achieved the best Root Mean Square Errors of Prediction (RMSEP) in most scenarios.
• Optimal performance was observed even with limited calibration set sizes.

Conclusions:

• TEAM is a robust and effective method for NIR spectra standardization.
• The proposed TEAM algorithm offers significant advantages over conventional methods, particularly in reducing prediction errors.
• The open-source implementation of TEAM in Python facilitates its adoption and further research.