Evaluation of driven piles’ load capacity by optimization-based prediction algorithms

In geotechnical engineering, a variety of models—both theoretical and empirical—have been put forward to calculate the load capacity of deep footings. In the present instance, physical hypotheses and the construction of estimations utilizing mathematical concepts are the main driving forces behind the models. The research findings are directly applicable to geotechnical engineering practice. Engineers and practitioners involved in foundation design, construction, and evaluation can use the developed models to estimate the load capacity of driven piles more accurately. The load capacity of driven piles (Q_t) using an experimental dataset has been estimated in the current essay using two-hybrid radial basis function ( \(RBF\) $\mathit{RBF}$ ) and multilayer perceptron ( \(MLP\) $\mathit{MLP}$ ) neural networks optimized with an arithmetic optimization algorithm ( \(AOA\) $\mathit{AOA}$ ). To ameliorate the modeling of the best network efficiency, the AOA method was used to identify the main components of each model (ARBF and AMLP). By raising the value of R² from 0.9653 to 0.9931 for the train portion as well as from 0.9516 to 0.9884 for the test portion, with increases of around 2.78% as well as 3.68% each, the AMLP model outperforms the ARBF network among the two constructed systems. The comparison between the findings of the outperformed model and the literature depicts a great improvement in the performance by raising R² from 0.96 to 0.9931. After evaluating the dependability and taking into account the assumptions, it is clear that the MLP coupled with AOA may score better than the RBF, that is the suggested one in Q_t prediction framework.