<p>Accurate prediction of wastewater treatment plant efficiency is vital for effective process management and operational optimization. This study models the performance of the South Tehran Wastewater Treatment Plant (STWWTP) using multiple machine learning methods. Six methods, including Artificial Neural Network optimized with Adam (ANN-GD), Genetic Algorithm–optimized (ANN–GA), Support Vector Machines (SVM), Decision Tree (DT), Random Forest (RF) and Extreme Gradient Boosting (XGBoost), were applied. Historical influent and effluent data from 2014 to 2022, covering ten key physicochemical parameters, were used for model development. Five parameters, including BOD, COD, TSS, pH and temperature measured at the plant inlet and operational stages, were used as model inputs, while effluent variables were used solely as prediction targets to predict effluent concentrations of BOD_out, COD_out and TSS_out. Model evaluation performed on unseen test data showed that XGBoost outperformed the other models and provided the highest predictive accuracy, with R², RMSE and MAE values (0.933, 3.570, 2.256) for BOD_out, (0.908, 14.25, 9.166) for COD_out and (0.936, 9.152, 5.102) for TSS_out. The results of this study show that the XGBoost model achieved high predictive accuracy, indicating its effectiveness for modeling, predicting and enhancing the operation of the South Tehran Wastewater Treatment Plant. The application of machine learning models can enhance the understanding and accurate prediction of wastewater treatment plant efficiency, offering valuable insights for WWTP management and optimization in arid and semi-arid regions.</p>