Data-Driven Prediction of Maximum Settlement in Pipe Piles under Seismic Loads

The structural stability of pipe pile foundations under seismic loading stands as a critical concern, demanding an accurate assessment of the maximum settlement. Traditionally, this task has been addressed through complex numerical modeling, accounting for the complicated interaction between soil and pile structures. Although significant progress has been made in machine learning, there remains a critical demand for data-driven models that can predict these parameters without depending on numerical simulations. This study aims to bridge the disparity between conventional analytical approaches and modern data-driven methodologies, with the objective of improving the precision and efficiency of settlement predictions. The results carry substantial implications for the marine engineering field, providing valuable perspectives to optimize the design and performance of pipe pile foundations in marine environments. This approach notably reduces the dependence on numerical simulations, enhancing the efficiency and accuracy of the prediction process. Thus, this study integrates Random Forest (RF) models to estimate the maximum pile settlement under seismic loading conditions, significantly supporting the reliability of the previously proposed methodology. The models presented in this research are established using seven key input variables, including the corrected SPT test blow count (N1)60, pile length (L), soil Young's modulus (E), soil relative density (Dr), friction angle (phi), soil unit weight (gamma), and peak ground acceleration (PGA). The findings of this study confirm the high precision and generalizability of the developed data-driven RF approach for seismic settlement prediction compared to traditional simulation methods, establishing it as an efficient and viable alternative.