Prediction of bearing capacity of pile using support vector and catboost regression.

The traditional techniques used for the estimation of the bearing capacity of piles are time consuming and costly. Moreover, some of the techniques are empirical, which involve enormous approximations and don't consider all the variables contributing to the strength of the pile. Hence, researchers have grown more interested in several machine learning approaches in order to consider more variables to model practical field situations with great accuracy and overcome the issues of approximation. In this study, cohesion, friction angle, specific weight of soil, pile-soil friction angle, flap number, pile area, and pile length have been considered as the contributing factors of pile-bearing capacity. Two of the most widely used machine learning models, Support Vector Regression (SVR) and CatBoost Regression, have been built with different architectures and hyperparameter sets for predicting the bearing capacity of pile. Datasets have been collected from various regions for modelling the heterogeneous nature of the soil and avoiding the overfitting issue to make the model more generalized. The R2 values have been found as 0.87 and 0.95 for concrete and steel piles respectively from SVR and 0.98 and 0.99 for concrete and steel piles respectively from CatBoost Regression. For the further improvement of the prediction accuracy some advance machine learning and deep learning models with more generalized dataset have been suggested. [ABSTRACT FROM AUTHOR]