SPT-BASED SOIL-LIQUEFACTION MODELS USING NONLINEAR REGRESSION ANALYSIS AND ARTIFICIAL INTELLIGENCE TECHNIQUES

Saturated, cohesionless soils can temporarily lose their shear strength due to increased pore-water pressure under the effect of repetitive dynamic loads such as earthquakes. This event is defined as soil liquefaction and causes significant damage to structures. the liquefaction potential of soils depends on many soil parameters obtained in the field and from laboratory tests. In this study new models have been developed to estimate the liquefaction potential of cohesionless soils. For this purpose, 837 soil data sets were collected to calculate the liquefaction potential with nonlinear multiple regression and artificial intelligence in the cities of Kayseri and Erzincan. the models based on Nonlinear Multiple Regression Analysis, Artificial Neural Networks, and Adaptive Neuro-Fuzzy-Inference System techniques were compared with the results of the simplified method. Determination coefficients (R2) and various error rates were calculated for the performance-evaluation criteria of the models. The proposed ANN model effectively found the complex relationship between the soil and the input parameters and predicts the liquefaction potential more accurately than other methods. It has an overall success rate of 90 percent and the lowest mean absolute error rate of 0.024. With the improvement of existing methods, new models have been introduced to estimate the liquefaction probability of soils.