Exploring Feature Extraction Methods for Raman Spectroscopy: A Comparative Study

Abstract

Background

Raman spectroscopy is a robust, non-destructive analytical technique that offers detailed insights into the chemical composition, molecular structure, and interactions of materials. However, the high-dimensional and complex nature of Raman spectral data requires effective feature extraction methods to reduce data volume and improve analysis. Efficient feature extraction methods are essential to reduce dimensionality while preserving critical spectral information. This study investigates and compares four feature extraction techniques, Principal Component Analysis (PCA), Independent Component Analysis (ICA), Multivariate Curve Resolution (MCR), and Non-negative Matrix Factorization (NMF), in the context of Raman spectroscopy to assess their ability to reduce the dimensionality of high-dimensional spectral data while preserving critical chemical and biological information.

Results

Using simulated datasets and real bacterial Raman spectra, we assessed how each method transformed high-dimensional Raman spectra into a lower-dimensional space, focusing on the structure of the reduced representations, their interpretability in terms of chemical and biological meaning, and their effectiveness in classification tasks. PCA and ICA effectively reduced dimensionality with minimal reconstruction errors, but they produced less interpretable features due to the orthogonality and independence constraints. In contrast, MCR and NMF generated chemically meaningful features and achieved classification performance comparable to that of PCA, even with fewer components.

Significance

For the first time, this study provides a systematic and in-depth examination of the reduced feature spaces generated by each method, offering a comprehensive understanding of their structural properties, interpretability, and impact on classification performance. The results highlight MCR as a particularly promising method for feature extraction in Raman spectroscopy. Its ability to produce chemically interpretable features and incorporate physicochemical constraints offers a key advantage over conventional techniques such as PCA. These characteristics make MCR especially suitable for analyzing complex Raman spectral data, where both classification accuracy and interpretability are crucial.