Changjing Zhou

Raman spectroscopy is a non-destructive analytical technique based on molecular vibrational properties. However, its practical application is often challenged by weak scattering signals, complex spectra, and the high-dimensional nature of the data, which complicates accurate interpretation. Traditional chemometric methods are limited in handling complex, nonlinear Raman data and rely on tedious, expert-knowledge-based feature engineering. The fusion of data-driven Machine Learning (ML) and Deep Learning (DL) methods offers a robust solution, enabling the automatic learning of complex features from raw data and achieving high-accuracy classification and prediction. The present study employed a structured narrative review methodology to capture the research progress, current trends, and future directions in the field of ML-assisted Raman spectral classification. This review provides a comprehensive overview of the application of traditional ML models and advanced DL architectures in Raman spectral analysis. It highlights the latest applications of this technology across several key domains, including biomedical diagnostics, food safety and authentication, mineralogical classification, and plastic and microplastic identification. Despite recent progress, several challenges remain: limited training data, weak cross-dataset generalization, poor reproducibility, and limited interpretability of deep models. We also outline practical directions for future research.