ABSTRACT
Liquefaction poses significant challenges in geotechnical engineering, particularly for foundation stability during seismic events. Conventional methods for assessing liquefaction rely on in-situ testing, which is inherently site-specific and lacks broader applicability. This study applies advanced ensemble-based machine learning (ML) techniques on laboratory-based cyclic triaxial test data to predict the cyclic resistance ratio (CRR) and assessing liquefaction susceptibility. Among the models tested, the Random Forest (RF) achieved the highest accuracy (R² = 0.93) with minimal error. Extra Trees (ET), Bagging Regressor (BR), and Gradient Boosting (GB) also performed well. Sensitivity analysis using RF identified relative density, mean particle size, and fines content as key factors affecting CRR. Validation against experimental data from sites in Italy and Japan showed strong CRR prediction accuracy. RF-predicted CRR is also used to calculate the factor of safety against liquefaction (FSL), closely matching field reports. The results demonstrate effectiveness of ML in improving liquefaction susceptibility analysis.
