Hybrid learning techniques for estimating bearing capacity of caisson foundation under general load V-H-M in natural clays.

Accurate prediction of the failure response of shallow caisson foundations under combined vertical, horizontal, and rotational loads is crucial in offshore geotechnical engineering. Previously, the authors developed a numerical framework using finite element limit analysis and a preliminary machine learning model trained with the Sand Cat Swarm Optimization (SCSO) algorithm. However, this earlier model lacked comparative evaluation with other optimization methods. This study expands upon that framework by assessing the performance of several nature-inspired optimization algorithms in training neural networks to predict failure envelopes. Using a dataset derived from high-fidelity simulations, the Whale Optimization Algorithm, Sand Cat Swarm Optimization, Secretary Bird Optimization Algorithm, and Egret Swarm Optimization Algorithm are employed to calibrate the models. Results indicate that the model using SBOA-trained model achieved the lowest test error. These findings underscore the potential of metaheuristic techniques in improving machine learning for offshore foundation design, offering a practical computationally efficient alternative.