Prediction of the uplift capacity of short bored piles in cohesive soil using artificial intelligence algorithms

Abstract

Estimating of the uplift capacity of short bored piles in cohesive soil represents a significant challenge in geotechnical engineering due to the complexity of soil–pile interaction and the uncertainties associated with traditional calculation methods. The main objective of this study is to evaluate and compare the performance of different machine learning algorithms for predicting uplift capacity using a database generated with the Finite Element Method. The study includes the numerical modeling of piles subjected to tensile loading in cohesive soil performed using PLAXIS 2D with an axisymmetric approach, as well as the development, training, and validation of predictive models within the KNIME Analytics Platform employing statistical metrics and both cross-validation and external validation schemes. The algorithms evaluated in this study include Linear Regression, Artificial Neural Networks, Decision Tree, Random Forest and Tree Ensemble models. The results show that machine learning models accurately reproduce the numerical response of the soil–pile system, with Artificial Neural Networks achieving the best performance, reaching an accuracy of 99%. These findings clearly confirm the potential of combining numerical modeling and machine learning as both an efficient and reliable tool for the analysis of deep foundations.