Prediction of soil liquefaction susceptibility using Artificial Intelligence

Liquefaction refers to the process by which a saturated granular material loses its strength and stiffness. This paper presents an example that demonstrates the role of neural networks in enhancing data synthesis to support empirical design development. The model applied to predict soil liquefaction susceptibility using data collected during several earthquake events. These data comprise a total of 210 case histories, including both liquefaction and non-liquefaction event. The machine learning technique employs a Multilayer Perceptron (MLP)-type neural network model (ANN). An MLP contains a number of layers of neurons; they employ nonlinear mapping mechanisms to model complex relationships and are, therefore, powerful tools for classification, regression, and pattern detection. The paper also highlights the potential of neural networks, the current model provides a sound and interpretable solution for the preliminary liquefaction screening on the basis of sparse but easily accessed data, and constitutes a suitable foundation for the further optimization with feature engineering and augmented datasets.