Evaluation of Liquefaction Potential Using Random Forest Method and Shear Wave Velocity Results

One of the great threats of earthquakes is soil liquefaction which can put many lives in danger and damage important structures. The evaluation of soil liquefaction potential is of great importance for geotechnical engineers in order to avoid future costs. The most common approach for such an evaluation has been using historical records of soil liquefaction, gathered from different locations. Most of the historical records include SPT and CPT data. This paper focuses on the evaluation of liquefaction potential using shear wave velocity (SVW) database, including 415 site observations. Among many models for liquefaction potential assessment, ANN and SVM are popular and used by different researchers to better predict the future hazards of liquefaction. In this paper, we introduce Random Forest as a method for liquefaction assessment to classify soils with respect to their potential of liquefaction when exposed to certain earthquakes. The details of feature selection and model development is stated. It is shown that Random Forest, tested on 83 unseen cases (20% of database), is able to predict the liquefaction potential in soils with 92.77% accuracy. Also, Random Forest can provide the relative importance of the predictors, among which the normalized shear wave velocity (V-s1) obtained the highest significance score for liquefaction potential prediction.