Multivariate Adaptive Regression Splines for Stability Number of Unsupported Conical Slopes in Anisotropic and Heterogeneous Clays

Abstract

AbstractThe novel equation for determining the stability number of unsupported conical slopes in anisotropic and heterogeneous clays is presented in this study. Multivariate Adaptive Regression Splines (MARS) model ~ a machine learning approach, is adopted to build the close–form equation between input variables and output results, and also investigate the coupling effects among input variables by sensitive analysis. The artificial data for MARS model is 576 combinations of four input variables (i.e., the ratio between the height and the radius at the bottom slope, and the inclination angle of slope, the gradient of increasing undrained shear strength, ratio of anisotropic) corresponding to output results of stability number which is based on finite element limit analysis (FELA) results from a previous study. The results of the paper provide a guidance theory and effective tool for practical engineering in determining the stability number of unsupported conical slopes in anisotropic and heterogeneous clays.KeywordsUnsupported conical slopesAnisotropic claysMARS modelArtificial intelligence and machine learning