Landslides pose significant risks in the mountainous Rif region of Morocco, particularly in the Oued Amter watershed, where prior studies rarely integrated advanced feature selection with machine learning to enhance prediction accuracy. This study employs Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) to develop a landslide susceptibility map, utilizing 1456 inventory points and identifying 13 key factors (stream, land use/land cover, elevation, and lithology), via Geographic Information System (GIS) analysis. The study confirms RF outperforms SVM and LR, achieving the highest accuracy (AUC: 0.93) compared to SVM (0.91) and LR (0.90), excelling in predicting high-risk zones. Influential factors include steep slopes (0°–75.90°) and elevation (185–7299 m), shaping susceptibility across in this semi-arid Mediterranean area. The study classifies the area into Susceptibility levels reveal, with a significant portion identified as high risk, with RF leading predictions. This study integrates feature selection techniques with Machine Learning Techniques (ML) models in a GIS framework, providing a robust approach for landslide risk mitigation and land-use planning. Machine learning’s integration with geospatial tools offers a cost-effective geohazard management solution, leveraging high-resolution data from Sentinel-2 and Google Earth. Future research could enhance models with real-time data, such as satellite rainfall estimates and ground deformation measurements, to improve temporal responsiveness and address socioeconomic vulnerabilities in Rif, Morocco.
Brahimi et al. (Thu,) studied this question.