Modeling rapidly discriminative strategies of Cr contaminated soils through machine learning

Abstract

Soil washing is employed to prevent the issue of Cr re-oxidation following the remediation of Cr-contaminated soil by transferring contaminants from the soil to the wash solution through the dissolving action. Nevertheless, rapidly screening effective washing agents remains challenging. This study conducted batch experiments to gather dataset on key factors (soil properties, Cr sequential extraction and type of washing agent) and Cr content after washing with 250 experimental data points. We developed 12 machine learning models, among which the light gradient boosting machine model excelled in predicting Cr content after washing, determining soil suitability for washing. Furthermore, the adaptive boosting and random forest models preferably predicted Cr content after various washing agents treatments, facilitating optimal detergent identification. Based on the feature analysis, soil pH, exchangeable potassium, reactive iron oxides, and Cr sequential extraction can account for most of the differences in washing treatment. To further validate our model, we used an additional 60 experimental data points, including soil properties with pH values beyond the range of the initial 250 data points. The R2 value of the predicted fit to the actual results was 0.801. This study provides the feasibility of improving the efficiency of technical remediation of Cr-contaminated soils.