Prediction of hydraulic conductivity of sodium bentonite GCLs by machine learning approaches

Six machine learning methods (linear regression, logistic regression, extreme gradient boosting (XGBoost), support vector machine, K-nearest neighbours and artificial neural network) were used to predict/classify the hydraulic conductivity of conventional sodium bentonite (Na-B) geosynthetic clay liners (GCLs) to saline solutions or leachates. Data were collected from the literature and randomly divided into two groups - that is, 80% of the data were used to train machine learning models and the rest, 20%, were applied to evaluate model performance. Features that are known to affect the hydraulic conductivity of Na-B GCLs (e.g. mass per unit area of GCLs, monovalent and divalent cations, ionic strength (I), relative abundance of monovalent to divalent cations (RMD), swell index and effective stress) were employed to predict/classify the hydraulic conductivity of Na-B GCLs. Comparative analyses were conducted with seven subsets corresponding to the combination of different features, and the best model was determined through cross-validation. The results showed that XGBoost consistently had the best performance among all methods over all subsets of features for both regression and classification analyses. Subset 4, using the swell index, I, RMD, I 2 x RMD, monovalent cations, divalent cations, effective stress and mass per unit area as features, outperformed all other six subsets in both regression analysis (R 2 = 0.826) and classification analysis (accuracy = 0.887) in the out-of-sample tests.