Modeling Static Liquefaction Susceptibility of Saturated Clayey Sand using Advanced Machine-Learning techniques

The identification of static liquefaction susceptibility is crucial for ensuring the safety and cost-effectiveness of structures. However, traditional estimation methods are often inefficient, time-consuming, and costly. In this study, a new alternative model was developed using ten advanced machine-learning methods, such as Deep Neural Network ( DNN ), Extreme Learning Machine ( ELM ), Support Vector Regression ( SVR ), LASSO regression ( LASSO ), Random Forest ( RF ), Ridge Regression ( Ridge ), Partial Least Square Regression ( PLSR ), Stepwise Regression ( Stepwise ), Kernel Ridge ( KRidge ), and Least Square Regression ( LSR ), to predict static liquefaction susceptibility in sands containing plastic fines. The model was trained on a dataset of 114 unconsolidated undrained triaxial shear tests implemented on saturated sand, and collected from the literature. Eight relevant factors were chosen based on literature recommendations as input parameters. The machine-learning methods were evaluated using six performance measures and K-fold cross-validation approach. The study found that the Deep Neural Network (DNN) model outperformed others, providing more accurate predictions and the closest to the experimental values. Finally, a reliable and easy-to-use graphical interface named "StaLique2024" was developed based on the DNN model. This latter will greatly offer a reliable and user-friendly graphical interface, facilitating efficient estimation of static liquefaction for researchers and civil engineers while saving time and money.