Forecasting water quality parameters using Random Forest and Gradient Boosting Machine regression models

Abstract

Mathematical modelling of the activated sludge process (ASP) enhances understanding of the process and improves effluent quality. However, as the process is complex and nonlinear, mathematical modelling has been challenging. In this study, Random Forest Regression (RFR) and Gradient Boosting Machine regression (GBM) are investigated and compared to predict biochemical oxygen demand (BOD), suspended Solids (SS) and pH for better control of wastewater treatment plants employing the activated sludge process. The study area selected was in a district of Kerala State in India. The model is evaluated using correlation coefficient R and the mean squared error (MSE). The software Python 3.11 is used for modelling. It was found that effluent BOD, SS and pH were predicted with maximum correlation coefficients of 0.6778, 0.9157 and 0.7578 and mean square errors of 0.005, 0.006 and 0.0102 respectively, by RFR. GBM improves the prediction of BOD, SS and pH, with correlation coefficients of 0.7346, 0.974 and 0.8778 and mean square errors of 0.0042, 0.0014 and 0.0051 respectively.