Prediction of Soil Liquefaction Using a Multi-Algorithm Technique: Stacking Ensemble Techniques and Bayesian Optimization

Abstract

Historically, liquefaction has caused a number of earthquake-related risks. When granular soils get saturated, liquefaction may occur during an earthquake, which can have devastating effects. Therefore, it is essential, especially in the context of civil and structural project planning, to have the capacity to precisely predict soil liquefaction potential. Therefore, the stacked ensemble-learning model with Bayesian optimization (BO-stacking) is introduced to make predictions of soil liquefaction more accurate. It was constructed utilizing primary algorithms like decision trees, support vector machines, and k-nearest neighbors, as well as secondary algorithms like the random forest algorithm. A Bayesian optimization method is also used to improve the accuracy of the predictions of soil liquefaction by adjusting the hyperparameters of these four classification algorithms. Information gain technique also was used for input selection. The results show that BO-stacking outperformed single prediction models. The testing accuracy and ACU of this model was 0.913 and 0.992, respectively. This study indicates that BO-stacking is a feasible alternative to established techniques for predicting soil liquefaction. In addition, the results of this study indicate that the BO and stacking approaches are effective in training the prediction model when used in conjunction.

Keywords

• earthquake
• soil liquefaction potential
• machine learning
• bayesian optimization
• stacked ensemble