Estimation of the Uplift Resistance for an Under-Reamed Pile in Dry Sand Using Machine Learning

Abstract Under-reamed piles are extensively used to resist uplift forces and settlements. The objective of the present study is to develop various machine learning models (linear and non-linear) and determine the best model to estimate the ultimate uplift resistance of under-reamed piles embedded in cohesionless soil. The machine learning models were developed considering the following input parameters: the density index, dry density, base diameter, angle of an enlarged base with a vertical axis, shaft diameter, and embedment ratio. A linear equation is proposed to estimate the ultimate uplift resistance based on Multivariate Linear Regression analysis with a mean absolute error equaling 0.25kN and 0.50kN for loose and dense sands respectively. The Decision Tree Regression model provides an excellent degree of accuracy with a mean absolute error of 0.02kN and 0.02kN in cases of loose and dense sands respectively.