The resilient modulus (Mr) of subgrade soils is an important element in road pavement structural design procedures. However, because of regional differences in soil parameters and the nature of test procedures, determining Mr in the laboratory has become cumbersome. The purpose of this study is to develop a reliable soft computing approach for forecasting Mr of subgrade soils using support vector machine‐based linear kernel (SVM‐LR), polynomial kernel (SVM‐Poly), radial basic function (SVM‐RBF), and random forest (RF) techniques. For this purpose, a dataset of 138 soil samples from diverse geographical locations in Ethiopia was gathered, and the models were trained and tested using an 85% training and 15% test data split. All of the recommended models used confining pressure (CP), notional maximum axial stress (NmaxAS), specific gravity, P#200, silt%, clay%, liquid limit (LL), plasticity index (PI), maximum dry density, Wopt, and swell as input variables. The correctness of the proposed models was assessed by comparing predicted and observed Mr values with various statistical approaches such as root means square error (RMSE), mean squared error (MSE), mean absolute error (MAE), and determination coefficient (R2). The support vector machine RBF model outperformed all other models in validation, with R2 values of 0.99 and 0.91, MAE of 0.1 and 1.76, MSE of 0.01 and 29.54, and RMSE of 0.1 and 5.43. The Taylor diagrams of the test and training datasets show that the SVM‐RBF model outperforms the other models. Sensitivity analysis was performed to explore the impact of input variable quantity on the output of Mr for the MPL model, and it was determined that the top five essential parameters are, in decreasing order, LL, P#200, silt%, PI, and clay%. In contrast, the nominal maximum axial stress and Gs have the least impact.
Read full abstract