Machine learning-enabled fluorescence spectrum source tracing of typical pollutants in industrial wastewater

Complex and difficult-to-trace pollutants in industrial wastewater are posing major challenges for regulation and treatment. The known source identification methods rely on rough water quality data and models, which limits the accuracy of source tracing. In this study, a convolutional neural network-principal component analysis-random forest (CNN-PCA-RF) fusion framework that directly utilizes the three-dimensional fluorescence excitation-emission matrix (EEM) data was established. An EEM database with five sect…

By the Truvace deskOctober 17, 20261 min read1 sourcePeer-reviewedTRV-2026-1744· cite & verify

Machine learning-enabled fluorescence spectrum source tracing of typical pollutants in industrial wastewater — "Sewer outlet" by eutrophication&hypoxia is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

In brief

Researchers built a fusion model that combines a two-dimensional convolutional neural network, principal component analysis and a random forest classifier. It reads three-dimensional fluorescence excitation-emission matrix data to work out where industrial wastewater came from. The training database covered five industrial sectors and 207 datasets.

Tracing complex pollutants is difficult for regulators and treatment operators, and current methods lean on coarse water quality data and models. A spectrum-based classifier could make source attribution more precise. The reported results come from a small database, so how the approach performs on real discharges remains open.

Main points

The CNN-PCA-RF framework uses raw three-dimensional fluorescence excitation-emission matrix data rather than coarse water quality indicators.
The database spans chemical, smelting, fine chemical, pesticide manufacturing and biotechnology sectors, with 207 datasets.
Trained on multi-concentration data, it reached 95.99% accuracy at the outlet level and 96.85% at the sector level.
On previously unseen random dilution factors, accuracy was 90.35% for outlets and 97.37% for sectors.

The gain

A machine learning framework working directly on 3D fluorescence spectra identified the source sector and outlet of industrial wastewater with roughly 90 to 97 percent accuracy in the study's tests.

The rundown

The paper says known source identification methods rely on rough water quality data and models, which limits tracing accuracy. The new framework addresses this by feeding spectral fingerprints straight into the classifier.

Each stage has a distinct role. The 2D-CNN extracts spatial-spectral features from the fluorescence matrices, PCA reduces their dimensionality, and the random forest provides stable classification.

Validation with random dilution factors the model had not seen during training tested whether it holds up when wastewater concentration changes. Outlet-level accuracy fell to 90.35%, while sector-level accuracy stayed at 97.37%.