Predicting Probability of Liquefaction Susceptibility Based on a Wide Range of CPT Data

In the present study, three efficient soft computing techniques (namely, genetic programming (GP), relevance vector machine (RVM), and multivariate regression splines (MARS)) are utilized to predict the probabilistic liquefaction susceptibility of soils based on reliability analysis. For this, a sum of 253 cone penetration test (CPT) data of 19 major earthquakes occurring between 1964 and 2011 have been collected from the literature. Six liquefaction parameters such as corrected cone penetration resistance, total vertical stress, total effective stress, maximum horizontal acceleration, magnitude moment, and depth of penetration are explored. To evaluate the overall performance of the proposed models, rank analysis has been carried out. Based on the values of performance indices, the GP model outperforms the other two models in terms of RMSE=0.15, R2 =0.77, and VAF=76.86 in the training stage while the same has been found 0.14, 0.81, and 80.46 in the testing phase. Also, the rank analysis confirms the superiority of the GP model in predicting the probability of liquefaction susceptibility of soils at all stages.