Machine Learning Applied to Predict Pile Bearing Capacity

Resumo

The present work presents the application of machine learning algorithms to the problem of predicting
load capacity of piles. A dataset was compiled from the literature, composed by 165 load tests associated with

SPT results performed in different regions of Brazil. From this raw base, 5 datasets were generated based on well-
known semi-empirical methods. Such sets were then applied to six ML algorithms and a linear regression. The

performances, measured by R2

and RMSE, were compared to those achieved by semi-empirical methods. The
RF technique stood out from the others, with a maximum R2 of 0.77. A case study was then carried out and its
results reinforced the good performance of ML algorithms against semi-empirical methods. Despite the limitations
of the work regarding the dataset, the conclusions point to the use of ML tools as a good alternative to the classical
methods of calculating load capacity.