Exploring feature extraction methods for Raman spectroscopy: A comparative study.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Raman spectroscopy provides detailed material insights but generates complex, high-dimensional data.
• Effective feature extraction is crucial for reducing data volume and improving analysis.
• This study compares four feature extraction techniques for Raman spectral data.

Purpose of the Study:

• To investigate and compare PCA, ICA, MCR, and NMF for feature extraction in Raman spectroscopy.
• To assess the dimensionality reduction capabilities of these methods.
• To evaluate the interpretability and classification performance of the extracted features.

Main Methods:

• Principal Component Analysis (PCA)
• Independent Component Analysis (ICA)
• Multivariate Curve Resolution (MCR)
• Non-negative Matrix Factorization (NMF)
• Analysis of simulated and real bacterial Raman spectra.

Main Results:

• PCA and ICA reduced dimensionality effectively but yielded less interpretable features.
• MCR and NMF generated chemically meaningful features.
• MCR and NMF achieved classification performance comparable to PCA, often with fewer components.

Conclusions:

• MCR is a promising method for Raman spectroscopy feature extraction due to its chemically interpretable features and physicochemical constraints.
• MCR offers advantages over PCA for complex spectral data analysis where interpretability and accuracy are key.
• This study systematically examines reduced feature spaces, providing insights into structural properties, interpretability, and classification impact.