Machine Learning Models for Predicting Shear Wave Velocity of Soils

As regards regions prone to seismic activity, shear wave velocity (V-s) is a design parameter for geotechnical systems exposed to dynamic loads. Evaluating V-s for geomaterials involves on-site and laboratory assessments; however, its availability is often limited in projects owing to resource and time constraints. Various mathematical and empirical models have been proposed to predict V-s for cohesive or granular soils; however, a majority of these models are specific to certain soil types and loading conditions. In this study, machine learning techniques were used for V-s prediction. These models encompass factors such as grading attributes, void ratio (e), mean effective confining pressure (sigma '(m)), consolidation stress ratio (KC), and specimen preparation methods. To achieve this, a series of bender element tests was performed on various sand and gravel mixtures supplemented with culled data from earlier investigations. This study facilitated the development of three machine learning models aimed at predicting the V-s for granular soils: artificial neural networks (ANN), support vector regression (SVR), and gradient boosting regression (GBR), aimed at predicting V-s for granular soils. The findings of the study demonstrated that the ANN model exhibited enhanced precision and reduced error compared with the other models.