Application of heavy metal immobilization in soil by biochar using machine learning

Biochar application is a promising strategy for the immobilization of heavy metal (HM)-contaminated soil, while it is always time-consuming and labor-intensive to clarify key influenced factors of soil HM immobilization by biochar. In this study, four machine learning algorithms, namely random forest (RF), support vector machine (SVR), Gradient boosting decision trees (GBDT), and Linear regression (LR) are employed to predict the HMimmobilization ratio. The RF was the best-performance ML model (Training R2 = 0.90, Testing R2 = 0.85, RMSE = 4.4, MAE = 2.18). The experiment verification based on the optimal RF model showed that the experiment verification was successful, as the results were comparable to the RF modeling results with a prediction error<20%. Shapley additive explanation and partial least squares path model method were used to identify the critical factors and direct and indirect effects of these features on the immobilization ratio. Furthermore, independent models of four HM (Cd, Cu, Pb, and Zn) also achieved better model prediction performance. Feature importance and interactions relationship of influenced factors for individual HM immobilization ratio was clarified. This work can provide a new insight for HM immobilization in soils.

Automated summary

Four machine learning algorithms were used to predict the immobilization ratio of heavy metals in soil, and the results were verified through experiments. This study provides new insights into the immobilization of heavy metals in soils.