Data-Driven Modeling of Peak Rotation and Tipping-Over Stability of Rocking Shallow Foundations Using Machine Learning Algorithms

The objective of this study is to develop data-driven predictive models for peak rotation and factor safety tipping-over failure rocking shallow foundations during earthquake loading using multiple nonlinear machine learning (ML) algorithms a supervised technique. Centrifuge shaking table experimental results on have been used the development k-nearest neighbors regression (KNN), support vector (SVR), random forest (RFR) models. input features ML include critical contact area ratio foundation; slenderness coefficient system; ground acceleration Arias intensity motion; categorical binary feature that separates sandy soil from clayey foundations. Based repeated k-fold cross validation tests models, we found overall average mean absolute percentage errors (MAPE) in predictions all three varied between 0.46 0.60, outperforming baseline multivariate linear model with corresponding MAPE 0.68 0.75. importance analysis reveals stability are more sensitive motion demand parameters than foundation capacity or type soil.