Estimation of California Bearing Ratio of stabilized soil with lime via considering multiple optimizers coupled by RBF neural network

As an extensively used experiment, the California Bearing Ratio (CBR) tests the resistance of soils in subgrade layers and superstructure foundations often used to design flexible pavements. Practically, since CBR tests are time-consuming and costly, only a limited number of them could be performed over a road construction project. In these cases, artificial-based prediction methods will be helpful as they are quick and cheap. Artificial neural networks (ANNs), including Radial Basis Function (RBF), are powerful tools in prediction procedures employing modeling philosophy. On the other hand, recently, because meta-heuristics are very efficient, academics have focused more on optimization utilizing them, reasonable execution time, and significant convergence acceleration rate in solving real-world problems. In this study, three different hybrid models are introduced comprising the neural network approach along with three optimizers [including adaptive opposition slime mold algorithm (AOSMA), gradient-based optimizer (GBO), and Sine cosine algorithm (SCA)]. Predicted values of CBR in two categories of training and testing models have been compared with measured values of CBR tests. Finally, through some evaluators, the efficiency of hybrid models was evaluated, and the best-proposed model was presented for practical applications. In addition, RBAO obtained the most suitable prediction values compared to other developed models.