Regressive approach for predicting bearing capacity of bored piles from cone penetration test data

In this study, the least square support vector machine (LSSVM) algorithm was applied to predicting the bearing capacity of bored piles embedded in sand and mixed soils. Pile geometry and cone penetration test (CPT) results were used as input variables for prediction of pile bearing capacity. The data used were collected from the existing literature and consisted of 50 case records. The application of LSSVM was carried out by dividing the data into three sets: a training set for learning the problem and obtaining a relationship between input variables and pile bearing capacity, and testing and validation sets for evaluation of the predictive and generalization ability of the obtained relationship. The predictions of pile bearing capacity by LSSVM were evaluated by comparing with experimental data and with those by traditional CPT-based methods and the gene expression programming (GEP) model. It was found that the LSSVM performs well with coefficient of determination, mean, and standard deviation equivalent to 0.99, 1.03, and 0.08, respectively, for the testing set, and 1, 1.04, and 0.11, respectively, for the validation set. The low values of the calculated mean squared error and mean absolute error indicated that the LSSVM was accurate in predicting the pile bearing capacity. The results of comparison also showed that the proposed algorithm predicted the pile bearing capacity more accurately than the traditional methods including the GEP model.