Rediction of Liquefaction in Sandy Soils Using Deep Learning Methods

This study investigates various deep learning algorithms such as fuzzy networks and k-nearest neighbors for predicting the liquefaction or non-liquefaction behavior of soil. Since soil liquefaction causes severe damage to infrastructures and lifelines, predicting this phenomenon is crucial. Two machine learning approaches were compared in this research to evaluate their effectiveness in predicting soil liquefaction. The models were constructed with multiple input parameters and a single output (liquefaction/non-liquefaction) under seismic conditions with a magnitude of ۷.۸. Model performance was assessed based on CPT (Cone Penetration Test) data using accuracy metrics in three states (liquefied, non-liquefied, and overall) along with confusion matrices and ROC (Receiver Operating Characteristic) curves. The study utilized models such as K-Nearest Neighbors (KNN) and fuzzy networks to evaluate soil liquefaction potential.

کلیدواژه ها:

Soil liquefaction ، Deep Learning ، K-Nearest Neighbor ، Fuzzy Network