Search

Machine learning was used to predict the effective diffusion coefficient of radionuclides in compacted bentonites to reduce the cost of experimental methods. Through-diffusion experiments were conducted to determine the effective diffusion coefficient of Re(VII), which was used as a surrogate for 99Tc(VII), in compacted Anji bentonite. Five parameters (the external surface area, the ionic strength, the mass ratio of montmorillonite, the compacted dry density, and the accessible porosity) that affect the effective diffusion coefficient were calculated by a multi-porosity model to generate data for the analysis of the machine learning models to overcome the limited experimental data. The effective diffusion coefficient was predicted using two popular machine learning models, the Light Gradient Boosting Machine and Artificial Neural Network models, where the former exhibited higher sensitivity and accuracy in the prediction than the latter. The performance of the machine learning models was validated by comparing the experimental effective diffusion coefficients between this study and previous studies. The present work revealed that the machine learning method can be a powerful tool and may offer a new means of studying the effective diffusion coefficient. • Effective diffusion coefficient was predicted by machine learning models. • Through-diffusion method measured the effective diffusion coefficient of Re(VII). • Dataset was generated by multi-porosity model. • Light Gradient Boosting Machine outperformed Artificial Neutral Network. [ABSTRACT FROM AUTHOR]