Numerical and machine learning-based assessment of seismic bearing capacity of footings adjacent to c - ϕ soil slope

Predicting the seismic bearing capacity of strip footings is challenging due to nonlinear soil-structure interaction, and neglecting the coupled effects of cohesion and unit weight often reduces accuracy for cohesive-frictional slopes. To address these shortcomings, this study integrates Lower Bound Finite Element Limit Analysis (LB-FELA) with Second Order Cone Programming (SOCP) and Modified Pseudo Dynamic (MPD) method to evaluate the seismic bearing capacity factor (Ncγs) for strip footings placed beyond the crest of a slope. The framework incorporates the effects of slope angle, normalised edge distance, non-dimensional cohesion factor, internal friction angle and horizontal seismic coefficient. Results demonstrate that Ncγs rises with greater cohesion, friction angle and increased distance from the slope edge, showing improvements when the footing is positioned sufficiently away from the crest. Seismic loading significantly lowers Ncγs, with reductions of 30–50% recorded for horizontal seismic coefficient, particularly in low-cohesion soils and steeper slopes. Gaussian Process Regression (GPR) and artificial neural network (ANN) were used to model nonlinear behaviour with uncertainty quantification and kernel evaluation. The GPR model achieves RMSE of 0.964 (training) and 1.562 (testing). A standalone executable programme enables rapid seismic bearing capacity estimation, reducing computation time by 95% for real-time applications.