Modeling the liquefaction potential of soils using different artificial intelligence techniques

Throughout history, earthquakes have caused great structural, functional and architectural damage. Structural damages, especially those that negatively affect the load-bearing system of the structure, lead to serious loss of life and property in such natural disasters. The degree of damage is related to the magnitude of the earthquake and earthquake parameters, as well as the behavior of the structures and the soils on which they are built. Therefore, for the reliability of a structure, it is of great importance to predict the behavior of the soil on which it is based during an earthquake. For this purpose, soil properties must be investigated in the best possible way and necessary engineering precautions must be taken against possible harmful effects. One of the most important causes of soil-related damage that may occur in structures is liquefaction. During soil liquefaction, small shear stresses that occur in the liquefied soil by earthquake action can lead to large deformations, that is, by completely changing the stress-strain behavior of the soil, causing significant damages such as bearing capacity losses, seismic settlements, flow failures, lateral spreading, retaining structure failures and sand boiling. In evaluating the liquefaction potential of soils, the results of the Standard Penetration Test (SPT) and Cone Penetration Test (CPT), which are the two most commonly used field and laboratory tests, are used. However, experiments both in the field and in the laboratory are very difficult and expensive procedures in terms of both taking undisturbed samples from the field, preparing the sample and bringing the sample to the laboratory environment to provide suitable conditions. For this reason, nowadays, numerical modeling methods, which lead to easier results, are preferred instead of such traditional methods, partly to overcome these difficulties, to save time, and partly to better model the soil behavior. In this thesis study, a data set consisting of past case records published in the literature, where earthquake characteristics, soil parameters and SPT-CPT test results are available for some regions where liquefaction is observed and not observed, has been compiled, and based on the information in this data set, Artificial Neural Networks (ANN), modeling studies were carried out with Adaptive Network-Based Fuzzy Inference System (ANFIS) and Data Mining Algorithms (DMA) programs, and the most effective parameter related to liquefaction was decided with the OneR algorithm in WEKA program and IBM SPSS software, which are among the statistical analysis methods. The analysis results were evaluated comparatively according to the coefficient of determination (R2) values, deviation angle D (°), mean absolute error (MAE) and root mean square error (RMSE) rates calculated using the data obtained from these models, and the correct classification percentages of the models when these three methods were compared to each other, it was seen that DMAs had a higher success rate in predicting liquefaction potential than ANN and ANFIS.