Predictive modeling and optimization of pharmaceutical pollutant adsorption on biochar using machine learning and data split analysis

In the present study, adsorption of pharmaceutical contaminants on biochar is examined by developing predictive machine learning models to estimate the adsorption capacity from the attributes of biochar and the experimental setup. For that purpose, a dataset comprising 2857 entries extracted from open-source research papers about emerging contaminants is used, where emphasis is placed on specific features, which include pyrolysis temperature, chemical elements, surface area, and adsorption conditions. Machine learning techniques, such as XGBoost, Random Forest, CatBoost, K-NN, Lasso Regression, Ridge Regression, and Linear Regression, are implemented for estimating the impact of these features on the adsorption capacity of drugs like ibuprofen, naproxen, atenolol, ethinylestradiol, diclofenac, and carbamazepine. Scatter and residual plots reveal the superior performance of tree-based models over traditional linear models with a high accuracy rate of 0.987. This research contributes to optimizing biochar design for remediation of pharmaceutical contaminants in aquatic environments.