Soft Computing Based Prediction of Ultimate Bearing Capacity of Footings Resting on Rock Masses

Abstract

The present study attempts to predict the ultimate bearing capacity of shallow foundations on rock masses using five different soft computing techniques namely i) random forest regression approach, ii) artificial neural networks, iii) support vector machine with polykernel, iv) support vector machine with (radial basis function) RBF kernel and v) M5P model tree. The ultimate bearing capacity was assumed to be dependent rock mass rating, unconfined compressive strength of rock, ratio of joint spacing to foundation width, and angle of internal friction for the rock mass. The performance of each model was analyzed by comparing the statistical performance measure parameters. The outcome of present study suggests that support vector machine (SVM) RBF kernel predicts the bearing capacity with least error followed by random forest regression model. All the model predictions further outperformed those based on empirical approach available in literature. Finally, sensitivity analysis performed for the SVM RBF kernel model suggests that the unconfined compressive strength of rock was an important parameter, in comparison to other parameters, considered for predicting the ultimate bearing capacity.