Optimizing the Efficiency of Water Pollution Tracing Based on Three-Dimensional Fluorescence Spectra Extracted from Characteristic Excitation Wavelengths

• • SVM model runtime reduced from 243.05 s to 34.56 s (86% decrease) while maintaining 94.4% recognition accuracy, enabling near-real-time water pollution tracing. • • Seven characteristic excitation wavelengths were selected via PARAFAC, correlation analysis, and feature importance analysis, reducing data dimensionality from full EEM scans without compromising classification performance. • • SVM outperformed random forest in both accuracy and efficiency, with precision, recall, and F1-score metrics confirming robust identification across six pollution source categories, especially metal surface processing wastewater. • • The method addresses the bottleneck of traditional EEM scanning (prolonged scan time and data redundancy) by using fixed excitation wavelengths, facilitating the development of compact, low-cost instruments for on-site applications.