Prediction of California bearing ratio using hybrid regression models

CBR assesses the subgrade strength of road infrastructure, typically requiring time-consuming laboratory testing. In this study, Machine Learning (ML) has been used to avoid the actual operation of laboratory tests of CBR. Support Vector Regression applies SVM principles as a supervised learning technique for predicting discrete values. To this aim, three hybrid models have been developed in this study, including SVWO, SVTE, and SVFD. To construct and verify the developed models, the study presented in this paper collected a dataset consisting of lime, lime slide, optimum moisture content, curing period, and Maximum Dry Density. Furthermore, several indicators, such as R², RMSE, MAE, MAPE, and U95, were working to compare and evaluate the hybrid models. According to the results, hybrid models are a viable option for forecasting CBR. Moreover, the study found that ML-based predictions are highly efficient in terms of time and energy consumption.