Prediction of soil liquefaction potential in central district of Erzincan province with artificial neural network

The specific physical, chemical and mineralogical properties of the soil cause different electricity productions. In such soils, liquefaction can often be observed under dynamic loadings such as earthquakes. Liquefaction potential is a situation that seriously disrupts the consumption of soils and the stability of structures. If the liquefaction in the soil cannot be predicted correctly, devastating damages may occur. For this reason, the prediction of liquefaction has become a remarkable research topic in the field of geotechnical engineering. Laboratory and field experiments are used to examine the liquefaction potential of soils. These performances may be insufficient due to human-induced errors that may occur in traditional programming methods. Due to this increase, the use of data-driven modern approaches of Artificial Neural Network (ANN) models is increasingly gaining importance. To benefit from the high accuracy prediction capacity of ANN models by learning from large data sets. The fact that Erzincan city center is a first-degree earthquake zone and many surrounding regions are exposed to severe earthquakes has caused it to be used as the target region for the study. This uses three different data sets to predict the liquefaction potential in the soil. General Observation Dataset (GGVK) was used in the training of the models, and both GGVK and Synthetic Dataset (SVK) were used in the testing phase to score the performance of the detailed model. With the obtained data, liquefaction predictions were made on the Target Area Dataset (HAVK) belonging to Erzincanşehir. In the study, Artificial Neural Network (ANN) and Polynomial Regression methods were used for liquefaction prediction, and ANN models used in Python and MATLAB environments were compared. The model effect of the input variables was analyzed with the permutation importance score method. The regression standard, testability accuracy and generalizability level were evaluated on both SVK and GGVK. In the last stage, regional liquefaction predictions were made by applying the data belonging to the formula HAVK and the areas with liquefaction risk were determined. The success of the model was evaluated with metrics such as F1-score and Accuracy Rate (Accuracy), and the obtained results reveal that the report models provide effective and reliable tools in estimating the liquefaction potential. As a result of the study, it was understood that the models, especially those whose applicability to different data sets was tested, provided strong and generalizable structures on a wide scale in engineering decisions.