A Comprehensive Assessment of Extreme Gradient Boost (Xgb) and Support Vector Regression (Svr) Performance to Predict Effluent Quality in Activated Sludge Treatment Process

Recent advancements in machine learning (ML) algorithms have shown that ML predictive models can be used as a rapid and efficient monitoring tool. This study proposes the capability of Support Vector Regression (SVR) and Extreme Gradient Boost (XGB) algorithms to predict effluent Total Suspended Solids (TSS) and Chemical Oxygen Demand (COD). In this regard, influent quality indicators and aeration basin parameters of a municipal wastewater treatment plant using activated sludge with extended aeration were considered as models' inputs, and data sets were divided into train, validate, and test data sets. In order to achieve the best model's input combination, several scenarios were considered, and the developed models' performances were compared to investigate their effect on the models' final accuracy. Each model's performance precision was determined by Mean Absolute Error (MAE), Mean Square Error (MSE), and Coefficient of Determination (R2). Results reveal that scenarios containing both groups of parameters helped the algorithms to improve final results in comparison to scenarios consisting of only influent indicators. Furthermore, both algorithms revealed acceptable accuracy. The MSE values of 6.452 and 6.778 for TSS prediction and 1.497 and 1.446 for COD prediction were achieved with XGB and SVR, respectively. However, XGB developed models with better performance results and more independence to models' inputs. Also, XGB showed high performance in predicting COD and TSS parameters for the unseen dataset with a confidence interval of 87%