Dual-channel surface-enhanced Raman spectroscopy integrated with machine learning for accurate classification of mixed dyes

Surface-enhanced Raman spectroscopy (SERS) has emerged as a powerful analytical tool for dye monitoring in environmental waters; however, challenges remain in resolving complex mixtures due to spectral overlap and limited signal reproducibility. To address this issue, a dual-channel SERS system integrated with machine learning (ML) was developed. Positively and negatively charged polyelectrolyte-modified Fe 3 O 4 @SiO 2 -Au substrates were employed for the selective detection of anionic and cationic dyes, respectively. Benefiting from the magnetic properties of the substrates, the system enables simultaneous and efficient capture of charge-contrasting dye mixtures. Several classification ML models, including linear discriminant analysis (LDA), k-nearest neighbor (KNN), and decision tree, were constructed to distinguish SERS data from various mixed dyes. The best-performing models, LDA and KNN, achieved classification accuracies exceeding 92 %, and a multimodel ensemble algorithm combining LDA and KNN was designed to harness the strengths of both models, resulting in a classification accuracy of 98 % for mixed dyes. This integrated approach enhances the efficiency and precision of Raman spectroscopic analysis while reducing the need for manual intervention, offering strong potential for high-precision pollutant detection and real-world application in analytical chemistry.