Environmentally informed prediction of ultimate bearing capacity using Kernel-SVM and PLAXIS 3D FEM for sustainable geotechnical design
Abstract
Accurately predicting Ultimate Bearing Capacity (UBC) is crucial for the safe and efficient design of foundations in geotechnical engineering. Yet, traditional methods remain time-consuming, costly, and environmentally disruptive. This study applies Support Vector Machine (SVM) models to develop a data-driven predictive framework for UBC estimation using optimized feature selection techniques and validated using experimental data and Finite Element Method (FEM). Three feature selection models (M1, M2, and M3) were evaluated, with M2 demonstrating the best predictive performance. Four SVM variants (Linear SVM, Quadratic SVM, Cubic SVM, and Cubic-SVM with Gaussian Kernel) were tested, with Quadratic SVM (QSVM) achieving the lowest Mean Absolute Error (MAE = 100.59) and Root Mean Square Error (RMSE = 113.55) in the testing phase, confirming its superior predictive accuracy. The results highlight the effectiveness of machine learning (ML) in reducing over-reliance on conservative design factors, optimizing material usage, and enhancing the sustainability of foundation engineering. Furthermore, ML-based geotechnical modeling offers environmental benefits, including minimized land disturbance, reduced excavation waste, and lower greenhouse gas emissions from construction activities. The study establishes QSVM with an optimized feature set as a reliable approach for UBC prediction, supporting climate-resilient infrastructure planning.