Machine learning–aided design of La-based composite modified biochar: Efficient materials and cost optimization for low-phosphorus water treatment

Abstract Phosphates are key contributors to eutrophication in water bodies. Lanthanum (La)-modified biochar (LaBC) offers notable advantages in achieving ultralow residual phosphate concentrations in water. However, the high cost of La limits its economic feasibility for practical use. This study applied machine learning (ML) models to optimize the design of La-based composite modified biochar, aiming to reduce application costs while maintaining effective phosphate removal to low residual levels. Eight ML models, namely random forest, gradient boosting regression (GBR), extreme gradient boosting (XGB), light gradient boosting, support vector machine, ridge regression, Bayesian ridge regression, and artificial neural network, were employed to predict the phosphate removal performance of La-based composite modified biochar. Results revealed that tree-based ensemble learning models (GBR and XGB: R 2 = 0.98 and 0.99, respectively) outperformed other models. Feature importance analysis indicated that adsorption reaction conditions and metal loading were the primary factors influencing residual phosphate concentrations. Experimental validation demonstrated strong agreement between actual removal efficiencies and model predictions. Based on actual phosphate concentrations in various lakes, economic costs, and treatment effectiveness, targeted material remediation strategies were proposed. The phosphate removal costs for La–Fe-modified biochar and two types of La–Ca-modified biochar were reduced by 59.25%, 55.10%, and 76.54%, respectively, compared with that of LaBC, achieving dual optimization of treatment effectiveness and economic cost. Overall, this study provides insights into developing low-cost, high-efficiency biochar materials and offers robust technical support for controlling water eutrophication. Graphical Abstract