Application of machine learning to the Vs-based soil liquefaction potential assessment

Earthquakes can cause violent liquefaction of the soil, resulting in unstable foundations that can cause serious damage to facilities such as buildings, roads, and dikes. This is a primary cause of major earthquake disasters. Therefore, the discrimination and prediction of earthquake-induced soil liquefaction has been a hot issue in geohazard research. The soil liquefaction assessment is an integral part of engineering practice. This paper evaluated a dataset of 435 seismic sand liquefaction events using machine learning algorithms. The dataset was analyzed using seven potential assessment parameters. Ten machine learning algorithms are evaluated for their ability to assess seismic sand liquefaction potential, including Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Naive Bayes (NB), K-Nearest Neighbor (KNN), Artificial Neural Network (ANN), Classification Tree (CT), Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM). A 10-fold cross-validation (CV) method was used in the modeling process to verify the predictive performance of the machine learning models. The final percentages of significant parameters that influenced the prediction results were obtained as Cyclic Stress Ratio (CSR) and Shear-Wave Velocity (V-S1) with 56% and 38%, respectively. The final machine learning algorithms identified as suitable for seismic sand liquefaction assessment were the CT, RF, XGBoost algorithms, with the RF algorithm performing best.