Machine learning-aided prediction of COD removal in the electrocoagulation process using a super learner model

A new predictive machine learning stacking model was developed to examine chemical oxygen demand (COD) removal efficiency in electrocoagulation. The model used a comprehensive dataset consisting of 379 points containing no missing data collected from different studies investigating COD removal efficiency using electrocoagulation, encompassing different wastewater types. The newly developed model included 10 input parameters, namely initial COD concentration, pH, conductivity, anode material, cathode material, inter-electrode distance, number of electrodes, current density, the ratio between the effective electrode area and the reactor volume, and electrolysis time. The stacking model uses three ensemble models, specifically, gradient boosting regression (GBR), eXtreme Gradient Boosting (XGB), and random forest regression (RFR), as the base learners, while the meta learner is a linear regression model. The developed model has a prediction accuracy of 95.3% for the R 2 value in the test dataset. Additionally, the study used sensitivity analysis and Partial Dependence Plots (PDPs) to determine the impact of each input parameter on COD removal efficiency. The results show that the three most influential parameters are electrolysis time, inter-electrode distance, and current density. • A comprehensive dataset of 379 COD removal results was compiled from electrocoagulation studies. • The dataset considered 10 distinct input parameters and was free of missing data. • The stacking model was developed using 3 ensemble models with linear regression as a meta-learner. • The stacking (super learner) model exhibited high prediction capacity with an R 2 of 95.3% in the test dataset. • The super learner model was interpreted using partial dependence plots and sensitivity analysis. • The most impactful parameters on COD removal were electrolysis time, inter-electrode distance, and current density.