Prediction of the Adsorption Behaviors of Radionuclides onto Bentonites Using a Machine Learning Method

This study builds a model to predict distribution coefficients (Kd) using the random forest (RF) method and machine learning based on Japan Atomic Energy Agency Sorption Database (JAEA-SDB). A database of ten input variables, including coefficient, pH, initial radionuclide concentrations, solid–liquid ratio, ionic strength, oxidation number, cation exchange capacity, surface area, electronegativity, radius, was constructed used for RF calculation. The calculation parameters employed in this work included two different hyperparameters, number decision trees maximum variables divide each node, together with seeds inside model. determination were derived various combinations hyperparameters seeds, assess result. Based results model, 22 target nuclides (Am, Ac, Co, Cm, Cd, Cs, Cu, Na, Np, Ni, Nb, U, Sr, Sn, Pb, Pa, Pu, Po, I, Tc, Th, Zr) predicted successfully. Among pH found make highest contribution determining coefficient. novelty lies first application predicting Kd value bentonites, JAEA-SDB. has established reliably coefficient radionuclides that is intended use evaluating arbitrary aqueous conditions.