Pseudo-static bearing capacity of skirted footings on sandy slopes using artificial neural network and random forest regression models

This study investigates the pseudo-static bearing capacity of skirted strip footings on cohesionless slopes using finite-element limit analysis and data-driven prediction models. A total of 216 numerical simulations were performed by varying soil strength, slope angle, seismic coefficient, footing location and skirt depth. The results showed that the inclusion of vertical skirts significantly enhances footing performance under seismic loading. However, increasing the horizontal seismic coefficient from 0 to 0.4 caused a considerable reduction in bearing capacity. Artificial neural network (ANN) and random forest regression (RFR) models were developed using 70% of the data set for training and 30% for testing. Both models achieved high prediction accuracy with coefficient of determination (R²) values greater than 0.90. The ANN model outperformed the RFR model, achieving a maximum R² value of 0.97. Sensitivity analysis indicated that skirt depth and width of the footing are the most influential parameter, contributing approximately 50.49% to the overall footing response. The proposed models provide a rapid and reliable approach for estimating the seismic bearing capacity of skirted foundations on sandy slopes.