Prediction of pile-bearing capacity using Least Square Support Vector Regression: individual and hybrid models development

The primary determinant in pile foundation design is the pile-bearing capacity (PBC), which relies on various soil characteristics and multiple parameters related to both soil and foundation. Accurately predicting PBC is crucial for structural safety, but traditional estimation methods often lack precision due to oversimplified assumptions. In response to this challenge, Machine Learning (ML) models have become formidable tools with sophisticated algorithms and extensive datasets. This research aims to develop predictive ML models using the Least Square Support Vector Regression (LSSVR) technique to estimate the PBC. These predictive models were developed by integrating LSSVR with optimization algorithms such as Prairie Dog Optimization (PDO) and Aquila Optimizer (AO), resulting in the LSSVR + PDO (LSPD) and LSSVR + AO (LSAO) models. A comprehensive database, including diverse pile characteristics and soil attributes obtained from literature sources, was employed to train and validate these models. The modeling results indicated that the LSPD model exhibited superior performance in terms of accuracy and reliability compared to the LSAO and LSSV models. It demonstrated a substantial improvement, with a 0.6% and 2% increase in R² and a notable decrease of 19.6% and 42.3% in RMSE compared to LSAO and LSSV, respectively.