Comparison of Accuracy in Prediction of Radial Strain in Stone Columns Using AI Based Models

Ground improvement of soft soil with construction of stone columns has been widely adopted. Lateral deformation of stone columns plays a significant role in behavior of columns. This study aims to explore the applicability of different AI techniques/mathematical models in predicting radial strain (ε) (change in radius/original radius of column) in stone columns as a function of significant input parameters viz. diameter (d) of stone column, l/d ratio, s/d (spacing/diameter) ratio, area ratio (A_r), λ (area of stone column/total area of loading), geosynthetic stiffness (k), β (clearance ratio). The radial strain (ε) in ordinary and encased columns is predicted with the help of linear regression, SVM, GPR and ANN models using Matlab software. The datasets of input parameters are obtained from already published literature. The values predicted by the models are compared to the corresponding true values of radial strain reported in the literature. A comparative analysis of the efficiency of all models is examined in terms of RMSE, R-squared, MSE and MAE values. It was observed that ANN models closely predicted the radial strain in columns with higher accuracy as compared to other models. ANN models may therefore be used to predict radial strain even in larger size columns in the field/in-situ conditions. However, these models are put forward as a complementary technique to evaluate the radial strain in columns and not as a substitute to field tests.