Predicting Wastewater Treatment Plant Effluent Quality Using Ensemble Learning

With the development of ensemble machine learning (ML) algorithms, complicated systems with nonlinear behaviors can be feasibly predicted in a cost-effective and time-saving manner. This study investigates the effectiveness of the ensemble ML tools for the estimation of quality indexes of wastewater such as chemical oxygen demand (COD), biochemical oxygen demand (BOD), pH, total suspended solids (TSS), total nitrogen (TN) and total phosphorus (TP). Water effluent discharged to Urmia Lake is of primary importance since it is a hypersaline lake that can negatively affect the composition and ecosystem of the lake. Therefore, ensemble methods are a very effective means of predicting and ultimately controlling the effluent discharged into the lake. Here, we examined Linear Regression (LR), Decision Tree (DT), and various ensemble ML algorithms including Random Forest (RF), Gradient Boosting Regressor (GBR), and Adaptive Boost (AdaB) Regression models, and evaluated their performance in the prediction of the parameters through models&rsquo metrics such as mean absolute error (MAE). According to the assessment results, AdaB outperforms its peers in anticipating COD (MAE = 2.76 mg/L), BOD (MAE = 2.64 mg/L), and pH (MAE = 0.14). Meanwhile, DT, RF, and GBR provided the most accurate estimates for TSS (MAE = 4.83 mg/L), TN (MAE = 1.94 mg/L), and TP (MAE = 0.24 mg/L), respectively. This study proves that ML methods can predict water quality indices accurately and efficiently based on previous experimental data thereby reducing reduce costs, labor effort, and hazardous experiences.