GIS and Machine Learning Framework for Landslide Susceptibility Mapping: A Raster-Based Approach
Year: 2026
Type: article
Abstract: Traditional landslide susceptibility models often rely on statistical methods applied to limited point-based samples, which can limit the spatial continuity of predictions. This study proposes a GIS-integrated, pixel-based machine learning approach that operates directly on raster data to generate continuous landslide susceptibility maps. Implemented in Rwanda, the workflow uses variables derived from a 30 m-resolution digital elevation model (DEM), including slope, aspect, elevation, and curvature, together with auxiliary environmental data such as NDVI, soil type, and distances to hydrological and infrastructural features. Several machine learning algorithms were evaluated, including Random Forest, Support Vector Machine, Logistic Regression, and ensemble approaches. Using only topographic variables, the Stacked Ensemble achieved the highest performance (F1-score = 0.8515; ROC-AUC = 0.9200). When all predictor variables were included, CatBoost achieved the highest performance (F1-score = 0.8860; ROC-AUC = 0.9564). This scalable approach is particularly suitable for landslide susceptibility assessment in data-scarce regions and can contribute to spatial planning and risk reduction efforts in Rwanda and similar regions.
Authors: Albert Iradukunda, Abdulkadir Kaan Özgün
Institution: Özyeğin University
Language: Turkish
Subfield: Management, Monitoring, Policy and Law
Field: Environmental Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7214227863