Application of machine learning for predicting ground surface settlement beneath road embankments

Predicting the maximum ground surface settlement (MGS) beneath road embankments is crucial for safe operation, particularly on soft foundation soils. Despite having been explored to some extent, this problem still has not been solved due to its inherent complexity and many effective factors. This study applied support vector machines (SVM) and articial neural networks (ANN) to predict MGS. A total of four kernel functions are used to develop the SVM model, which is linear, polynomial, sigmoid, and Radial Basis Function (RBF). MGS was analysed using the nite element method (FEM) with three dimensionless variables: embankment height, applied surcharge, and side slope. In comparison to the other kernel functions, the Gaussian produced the most accurate results (MARE = 0.048, RMSE = 0.007). The SVM-RBF testing results are compared to those of the ANN presented in this study. As a result, SVM-RBF proved to be better than ANN when predicting MGS.