Industrial wastewater identification based on HPLC combined with data standardization and ensemble learning algorithms

Abstract

With the continuous development of industry, accidental or unauthorized discharges of wastewater from manufacturing enterprises have increasingly severe impacts on the environment. Rapid and accurate identification of industrial wastewater sources is of great significance for enhancing regulatory oversight and environmental protection. In this study, we propose a novel approach for discriminating the sources of industrial wastewater by integrating data standardization with ensemble learning. High-performance liquid chromatography (HPLC) is employed to collect wastewater samples. To ensure data consistency and accuracy, a local stretching alignment method combined with Gaussian fitting is introduced for precise peak alignment in chromatographic data. We compare the modeling performance of two ensemble learning algorithms: Random Forest (RF) and Extreme Gradient Boosting (XGBoost). To further improve model accuracy, hyperparameter optimization is conducted using the Optuna framework. The models are systematically evaluated through five-fold cross-validation. Experimental results show that the optimized RF model achieves an average cross-validation accuracy of 97.87%, a test set accuracy of 98.28%, and an F1 score of 0.9799, the accuracy on the newly collected samples reached 95.08%, demonstrating excellent overall performance and a well-balanced trade-off between precision and recall. This approach provides an efficient and reliable analytical tool for tracing the sources of industrial wastewater.