A number of studies have underscored the use of models developed via artificial neural networks (ANNs) in forecasting the bearing capacity of driven piles. Nonetheless, the main drawbacks of using the techniques relying on artificial neural networks are their slow convergence rate and reliable testing outputs. This investigation aims to present and evaluate four optimized artificial neural networks (ICA, IWO, PSO, and LCA) to determine the bearing capacity of driven piles deployed in cohesionless soils. The results of in situ studies are employed as the training data for optimization of the ICA-MLP, IWO-MLP, PSO-MLP, and LCA-MLP structure. The employed input parameters used for developing the models are the pile diameter (cm), pile length (L), undrained shear strength (kPa), and effective vertical stress (kPa). The models’ output is the ultimate bearing capacity of a driven pile deployed in cohesionless soils. In order to illustrate the capability of the hybrid models, a comparison is made between the predicted results and those obtained using a pre-developed ANN model. As such, the values obtained for the coefficient of determination (R2) for testing and training datasets of ICA-MLP, IWO-MLP, PSO-MLP, and LCA-MLP models were (0.933, 0.944, 0.959 and 0.955) and (0.989, 0.99, 0.978 and 0.987), respectively. Additionally, the variance values obtained for Root Mean Square Error (RMSE) for testing and training datasets of ICA-MLP, IWO-MLP, PSO-MLP, and LCA-MLP models were (0.168, 0.082, 0.078, and 0.129) and (0.055, 0.063, 0.092, and 0.063), respectively. The acquired results reflect that the developed model of IWO-MLP is of high reliability. One can present the model as a novel model used in deep foundation engineering.