Experimental Characterization-Based Machine Learning Modeling for the Estimation of Geotechnical Properties of Clay Liners

Expansive soils are widely used as compacted clay liner materials in landfills applications. The evaluation of these soils as contaminant barriers involves tedious experimental work which needs to be reduced by predictive computer-assisted approaches. This study evaluated the geotechnical properties of soil bentonite admixtures (with bentonite content of 0–16%) under the hydration of salt solutions. Advanced statistical and machine learning techniques were employed to predict geotechnical properties based on index tests. The laboratory experiments showed that the values of liquid limit, plastic limit, swell potential, swell pressure, and swell index of the tested soil samples increased while hydraulic conductivity decreased with the increase in bentonite percentage. The developed multivariable regression models for hydraulic conductivity and swell pressure were predicting the target values with R² values of 0.77 and 0.79 respectively. The corresponding RMSE values were 0.7 m/s and 2.51 kPa. In all the applied machine learning approaches, the Support Vector Machine Regression had the highest value of R² (0.95) and minimum value of RMSE (2.62 kPa) amongst the random forest, k-nearest neighbor, and support vector machine. In this study, advanced machine learning modeling was performed for the estimation of critical properties required for the successful implementation of clay barriers.