Bearing capacity prediction for spatially variable clay with rotated anisotropy using ANN-driven stochastic modeling

Shallow foundations are commonly used in geotechnical engineering, requiring accurate estimation of soil strength, particularly for undrained random clay with rotated anisotropy. This study employs a random adaptive stochastic analysis to assess the bearing capacity and failure probability of strip footings on such soils, considering key parameters including undrained shear strength ( S u ), its coefficient of variation ( COV Su ), spatial correlation lengths ( CL x and CL y ), and rotation angle ( β ). Results include the mean bearing capacity factor ( μN ran ), probability of failure ( PoF ), and failure mechanisms influenced by COV Su and CL y . Four hybrid ANN-based soft computing techniques, namely Ant Lion Optimizer (ALO), Imperialist Competitive Algorithm (ICA), Shuffled Complex Evolution Algorithm (SCE), and Teaching-Learning-Based Optimization (TLBO), were developed and evaluated using various performance metrics ( R², MAE, RMSE, VAF, IOS , and RSR ), as well as convergence curves, regression plot, Taylor diagram, ranking of model performance, and relative impact. Among them, the ANN-SCE model demonstrated superior performance, achieving R² values of 0.9980 (training) and 0.9905 (testing), with consistently low errors across all metrics. These findings provide a robust framework for predicting foundation behavior under spatially variable soil conditions and highlight the potential of hybrid AI models in enhancing geotechnical reliability and design efficiency.