Volume 2, Issue 1 — Year 2025 — Article e100028

Abbas Abgrami¹; Wu-Feng Zhang²; Hu Mao³; Li Wang⁴

1. School of Civil Engineering, Qingdao University of Technology, Qingdao 266520, China
2. School of Civil Engineering, Qingdao University of Technology, Qingdao 266520, China
3. School of Civil Engineering, Qingdao University of Technology, Qingdao 266520, China
4. School of Civil Engineering, Qingdao University of Technology, Qingdao 266520, China

This study presents a GIS-based comparative analysis for landslide susceptibility mapping in Kelardasht County, utilizing artificial neural networks (ANN), support vector machines (SVM), and random forest (RF) models. The study identifies key triggering factors, including slope angle, aspect, lithology, land-use, proximity to faults, proximity to rivers, proximity to cities, proximity to occurred landslides, rainfall, and elevation, and 42 historical landslide records. The methodology involves integrating spatial data into a GIS framework, applying data preprocessing and feature selection techniques, followed by training and validation of the ANN, SVM, and RF models. Model performance was assessed using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and the receiver operating characteristic (ROC) curve. The RF model demonstrated superior accuracy with the highest area under the curve (AUC) value, followed by ANN and SVM, indicating its robustness in identifying high-susceptibility zones. Verification results confirm the reliability of the models, providing precise susceptibility maps that classify the region into five risk categories: very high, high, moderate, low, and very low. The findings offer essential insights for regional planners and policymakers, enabling informed decisions on mitigation strategies and land-use planning to minimize landslide impacts. This comparative approach underscores the value of machine learning models in advancing landslide susceptibility assessment.

Landslide susceptibility, Machine learning, Geohazards, SVM, ANN