Soft computing for determining base resistance of super-long piles in soft soil: A coupled SPBO-XGBoost approach

Abstract

Prediction of base resistance for long piles is usually challenging because of the complex mobilization of load over the depth. This study hence proposes a novel machine learning (ML) method where the eXtreme Gradient Boosting (XGBoost) algorithm is coupled with the state-of-the-art Student Psychology-Based Optimization method (SPBO) to predict base resistance of piles. A dataset (796 datapoints) from well-instrumented static load tests among various projects in the soft soil region of Ho Chi Minh City is used to develop the model. Five input parameters including the applied load, loading point displacement, axial stiffness, SPT value of the soil beneath the pile toe, and the distance from the loading point to the pile toe are considered. Performance of the coupled SPBO-XGBoost model is comprehensively assessed by comparing it with other optimized XGBoost models including the Teaching-Learning Based Optimization, Bayesian Optimization, and other well-known ML techniques. The established model is used to predict data obtained from independent studies and compared with practice design codes of pile foundation. The results show that the proposed SPBO-XGBoost model outperforms all other models considering the prediction accuracy and reliability. Sensitivity analysis and Partial Dependence Plots are conducted to gain insights into the base resistance mechanism, giving considerable implications to design practice of pile foundation.