Classification of Seismic-Liquefaction Potential Using Friedman’s Stochastic Gradient Boosting Based on the Cone Penetration Test Data

The analysis of liquefaction potential of soil due to an earthquake is a classical problem for civil and geotechnical engineers. In this paper, Friedman’s stochastic gradient boosting (FSGB) method is introduced and investigated for the prediction of seismic liquefaction potential of soil based on the cone penetration test (CPT) data. The SGB models were developed and validated on a relatively large dataset comprising 226 field records of liquefaction performance and CPT measurements. The database contains the information about effective vertical stress, cone tip resistance, total vertical stress, sleeve friction ratio, depth of potentially liquefiable soil layer, earthquake magnitude and maximum horizontal ground surface acceleration. To find the most suitable model, several different combinations of above input parameters were tested to assess the usefulness of SGBs for liquefaction assessment using CPT data. SGBs are based on classification & regression trees with ensemble learning strategy and found to work well in comparison to artificial neural network and support vector machine models. The developed SGB provides a viable tool for practicing engineers to determine the liquefaction potential of soil.