This study investigated the undrained vertical bearing behaviour of circular skirted mudmat foundations resting on anisotropic marine clays using finite element limit analysis (FELA) incorporating Anisotropic Undrained Shear failure criterion. The effects of mudline undrained shear strength \(\:{s}_{u0}\) , strength gradient \(\:k\) , undrained strength anisotropy ratio \(\:{r}_{e}\) , and skirt depth ratio \(\:d/D\) on the ultimate bearing pressure and normalized bearing capacity factor \(\:{N}_{su}\) are systematically examined. The results reveal strong nonlinear interactions between skirt embedment, strength non-homogeneity, and anisotropy. \(\:{N}_{su}\) increased significantly with increasing \(\:d/D\) and \(\:k\) , especially under low \(\:{s}_{u0}\) , while higher \(\:{r}_{e}\) enhanced vertical resistance. The numerical database was used to train supervised regression models, including linear regression, support vector regression, regression trees, wide neural networks, and Gaussian process regression (GPR). Model performance was assessed using RMSE, MAE, MAPE, and \(\:{R}^{2}\) . Among these, GPR and wide neural network achieved the highest predictive accuracy, reproducing the strongly nonlinear FELA trends with \(\:{R}^{2}\) values approaching 0.99. Quadratic response surface models were also developed for discrete values of \(\:{s}_{u0}\) , providing closed-form expressions for \(\:{N}_{su}\) . The integrated FELA, machine learning, and response surface framework provided accurate predictive capability and practical formulations for assessing the vertical bearing capacity of skirted mudmat foundations within the investigated parameter range.
