Comparative analysis of supervised learning models for effluent quality prediction in wastewater treatment plants

Effluent quality prediction is critical for optimizing Wastewater Treatment Plant (WWTP) operations, ensuring regulatory compliance, and promoting environmental sustainability. This study evaluates the performance of five supervised learning models—AdaBoost, Backpropagation Neural Networks (BP-NN), Support Vector Machine (SVR), XGBoost, and Gradient Boosting (GB)—using data from a WWTP in Zhuhai, China. The Effluent Quality Index (EQI), integrating multiple pollutant concentrations and environmental impacts, was used as the target variable. The models were trained and tested on 84 monthly datasets, with their performances compared using R2, Mean Absolute Percentage Error (MAPE), and Mean Bias Error (MBE). XGBoost achieved the best balance between accuracy and robustness, with the lowest MAPE(6.11%) and a high R2(0.813), while SVR excelled in fitting accuracy(R2 = 0.826) but showed limitations in error control. Although we employed GridSearchCV with cross-validation to optimize hyperparameters and ensure a fair model comparison, the study is limited by the reliance on data from a single WWTP and the relatively small dataset size (84 records). Nevertheless, the findings provide valuable insights into selecting effective machine learning models for effluent quality prediction, supporting data-driven decision-making in wastewater management and advancing intelligent process optimization in WWTP.