| Abstrakt: |
One of the possible consequences of earthquakes in saturated areas is soil liquefaction and as a result the failure of foundations of buildings, types of infrastructure, bridges and many other disasters. In this study, in order to evaluate the potential of soil liquefaction on 79 samples from China Tangshan Earthquake Database, several intelligent classification models were constructed with the help of Orange software. Therefore, the performance of 5 intelligent classification methods (Logistic Regression, Artificial Neural Network (ANN), Support Vector Machine (SVM), K-fold Nearest Neighbor (KNN) and Random Forest) were compared based on different criteria. The results showed that SVM, ANN and Logistic Regression methods have a high ability to predict soil liquefaction class and among them the Logistic Regression method with AUC index (0.98) was selected as the best method. In addition, the study of the effectiveness of variables using three criteria of Information Gain, Information Gain Ratio and Gini Index, indicates that the variable measured CPT tip resistance is the most effective variable and is the first priority. The variables of cyclic stress ratio and peak acceleration at the ground surfaceare also important features. [ABSTRACT FROM AUTHOR] |
