The Use of Machine Learning Approach to Predict Pile Capacity in Non-Cohesive Soils

Existing theoretical and empirical pile design methods cannot accurately model the complex interaction between piles and soil. Consequently, there is a growing trend towards utilizing machine learning techniques to better capture the nonlinear soil-pile interaction. This paper aims to predict the capacity of bored piles in cohesionless soils using a machine learning approach. The machine learning algorithm was trained using a database of 18 bored pile cases in non-cohesive soils and validated with a separate dataset of 8 bored piles in cohesionless soil. Moreover, the performance of the machine learning method was compared with that of a traditional pile design method (i.e., SA-SPT method) in Southern Africa. The evaluation was based on the ratio of measured capacity to predicted capacity (Qm/Qp) statistics and the coefficient of determination (R2). The results showed an R2 of 0.89 for the machine learning method compared to 0.85 for the SA-SPT method, indicating the superior accuracy of the machine learning approach in predicting pile capacity.