Machine Learning-Enabled Intelligent Analysis of Surface-Enhanced Raman Scattering: Methods, Applications, and...

Surface-enhanced Raman spectroscopy (SERS) enables ultrasensitive molecular detection but produces high-dimensional and substrate-dependent spectral data that are difficult to analyze using conventional methods. The integration of machine learning (ML) provides new opportunities for extracting chemical information from complex SERS datasets and for optimizing nanostructured substrates that determine signal enhancement. This review summarizes recent advances in ML-assisted SERS across the analytical workflow. Data characteristics and preprocessing strategies are first outlined, followed by an overview of supervised, unsupervised, and deep learning approaches for spectral classification and quantitative analysis. Applications in biomarker discovery and spectral fingerprint recognition are discussed, with emphasis on model interpretability. In addition, ML-driven strategies for substrate optimization, including surrogate modeling and inverse design, are highlighted as emerging directions for improving enhancement efficiency. Current challenges, such as data scarcity, limited generalization, and real-time deployment constraints, are also examined. The convergence of ML and SERS is gradually shifting Raman-based analysis toward more predictive and integrated sensing frameworks.