Assessment of influential geotechnical and seismic features for soil liquefaction using SPT- and Vs-based machine learning models

This study presents a machine learning-based framework to evaluate soil liquefaction susceptibility using two geotechnical datasets: Standard Penetration Test (SPT) and Shear Wave Velocity (V s ), integrated with strong ground motion parameters. Feature selection techniques were applied to identify the most influential predictors. For the SPT dataset, corrected blow count, maximum acceleration, and velocity pulse period were identified as key factors. For the Vs dataset, shear wave velocity, epicentral distance, maximum displacement, and the number of effective cycles were found to be critical. Three classification models; Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Neural Networks (NNs) were developed and evaluated using accuracy, precision, recall, F1 score, and AUC (Area Under Curve). Among them, RF consistently achieved the best performance, with accuracy and AUC values of 0.78 and 0.83 for the SPT dataset, and 0.75 and 0.83 for V s dataset, respectively. The results demonstrate that combining feature selection with ensemble learning enhances the reliability and predictive performance of liquefaction susceptibility assessment across varying site conditions. This study presents a unified machine learning framework integrating multidimensional strong ground motion parameters with both SPT- and Vs-based liquefaction datasets. The results demonstrate that combining advanced seismic features with ensemble learning significantly enhances the reliability of liquefaction susceptibility prediction.