Sparse wavelengths data in mid-infrared spectroscopy: Modelling approaches and channel sampling

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Development of cost-effective infrared instruments for biological sample analysis.
• Sparse data, characterized by limited single-frequency or narrow multi-frequency channels, presents challenges for modeling.
• Minimizing channel count is crucial for reducing instrument costs.

Purpose of the Study:

• To investigate the impact of channel sampling on data modeling outcomes.
• To identify optimal modeling algorithms for various types of sparse infrared data.
• To evaluate regression techniques for predicting fatty acid composition from sparse spectral data.

Main Methods:

• Simulation of sparse Fourier Transform Infrared (FTIR) spectra from milk and fungi samples.
• Application of Partial Least Squares Regression (PLSR), Multiple Linear Regression (MLR), and Random Forest (RF) for predictive modeling.
• Evaluation of model performance based on varying numbers and types of spectral channels.

Main Results:

• PLSR demonstrated excellent performance for multi-frequency sparse data, achieving robust calibration models with as few as three channels.
• MLR and RF methods yielded comparable results for single-frequency sparse data, requiring a total of nine channels.
• The findings indicate that specific algorithms are well-suited for different sparse data structures.

Conclusions:

• PLSR is a highly effective algorithm for modeling multi-frequency sparse infrared spectral data.
• Optimal algorithm selection depends on the nature of the sparse data (single-frequency vs. multi-frequency).
• This research provides a pathway for developing accurate and cost-efficient spectroscopic analysis methods for biological samples.