Soft voting ensemble classifier for liquefaction prediction based on SPT data

Abstract

Soil liquefaction, caused by increased porewater pressure, is a significant risk in seismically active areas, impacting infrastructure stability and challenging liquefaction forecasting due to intricate nonlinear interactions. This study proposes a soft voting ensemble classifier (SVEC) that integrates CatBoost Classifier (CBC), Random Forest Classifier (RFC), and Gradient Boost Classifiers (GBC) to predict liquefaction using Standard Penetration Test (SPT) data. The dataset of 540 soil and seismic parameters was utilized to develop SVCE. The dataset incorporates depth, SPT-N60 values, Fine Content of soils (FC), Ground Water Table (GWT), Effective Stresses of Overburden (ESO), Total Stresses of Overburden (TSO), Earthquake magnitude (M_w), and Peak Ground Acceleration (PGA), as input factors for liquefaction prediction. The proposed model was evaluated through performance metrics (Accuracy, Recall, Precision, and F1-score), confusion matrix, sensitivity analysis, feature importance, and Shapley additive explanation (SHAP) analysis. SHAP improves the reliability of ensemble techniques in liquefaction analysis by highlighting the most critical input features, such as PGA, SPT-N60, FC, and GWT. tenfold cross-validation and precision-recall curve confirms the SVEC model’s robustness, achieving a high accuracy of 99.38% in accurately predicting liquefaction.