Prediction of Liquefaction-Induced Settlement Using Artificial Neural Network

This study aims to propose a machine-learning algorithm for predicting the ground settlement caused by liquefaction. An artificial neural network (ANN) approach was used. The properties of soil layers, namely unit weight (γ), soil layer depth (d), standard penetration test blow count (N1(60)), cyclic stress ratio (CSR), and corresponding settlements were selected to train, validate, and test the proposed model. Using the R-squared, the proposed model was compared to other machine learning models like linear regression, elastic net regression, polynomial regression, and support vector machine. For the comparison between the real and predicted settlements, the experimental results show that while the lowest R2 value of 0.322 was found from elastic net regression, the highest accuracy of 0.871 was obtained from the proposed ANN model. It concluded the effectiveness of the machine learning method, particularly in the ANN model, in predicting the soil characteristics.