Enzymatic remediation of bisphenol A from wastewater in the presence of bio- and chemical-based additives: A machine learning study

Phenolic wastewaters discharged from various industrial sectors pose an alarming risk to the human health and the ecosystem. Enzymatic remediation of phenolic wastewater, inspired by the catalytic function of laccases, has recently emerged as an attractive and environmentally benign treatment technology. However, this process is still expensive and laborious. Thus, the development of machine learning models for predicting the enzymatic remediation efficiency of phenolic pollutants would minimize the laborious experimentation efforts and save the resources. Accordingly, machine learning prediction of the enzymatic removal of bisphenol A (BPA), as a model phenolic pollutant, from wastewater samples has been explored in this work for the first time, to the best of our knowledge. The experimental data were collected under various conditions (i.e., different temperatures, pH, salinity, initial BPA concentration, and enzyme concentration) in the presence of bio-based (i.e., rhamnolipids) and chemical based (i.e., polyethylene glycol (PEG), Triton X-100, cetyltrimethylammonium bromide (CTAB), and sodium dodecylbenzenesulfonate (SDBS)) additives. The collected data were utilized to train various machine learning models. Among the utilized models (i.e., support vector regression (SVR), extreme gradient boosting (XGBoost), random forest, artificial neural network (ANN), linear regression, and K-nearest neighbors (KNN)), the XGBoost model has demonstrated superb predictive accuracy of the BPA removal efficiency with a coefficient of determination (R²) of 0.979 and a root mean square error (RMSE) of 3.171. To ensure reliability of the developed models, they were well-trained using about 80 % of the data and their parameters were carefully calibrated before applying them to the remaining unseen data during the testing phase.