University of Tabriz Journal of Geography and Planning 2008-8078 30 96 2026 08 23 Landslide Susceptibility Zoning Using MLR Modeling Technique: A Case Study of Deh-Sheikh Basin Landslide Susceptibility Zoning Using MLR Modeling Technique: A Case Study of Deh-Sheikh Basin 20233 10.22034/gp.2025.66379.3379 FA Mahshid Moavi PhD student at the Faculty of Geographical Sciences and Planning, University of Isfahan Heeva Elmizadeh Faculty of Khorramshahr University of Science and Technology Mojgan Entezari Mojgan Faculty of Geographical Sciences and Planning, University of Isfahan Journal Article 2025 03 12 This research aimed to map landslide susceptibility in the Dehsheikh watershed in northeastern Khuzestan province, identifying high-risk areas and proposing practical solutions using the Multiple Linear Regression (MLR) algorithm. The study examined 15 influential parameters across four main categories: geological factors (formations, faults, earthquakes), topographic factors (slope, aspect, elevation), environmental factors (precipitation, land use, vegetation cover), and hydrological factors (distance from rivers, Topographic Wetness Index (TWI), and SPI). In this study, 129 historical landslide points were identified and 2,500 random points were generated as control samples. The data were split into a 70:30 ratio for model training and testing. After preparing information layers in ArcGIS, SAGA-GIS, and ENVI software, modeling was performed using the MLR algorithm in the RStudio environment.<br /><br />The ROC curve validation showed the model had excellent performance in landslide prediction with AUC = 0.909 for training data and AUC = 0.906 for test data (p < 0.01). Factor analysis revealed the geological formations had the greatest impact on landslide occurrence (coefficient = 0.60, p < 0.001). Specifically, the Quaternary and Gachsaran formations showed the highest sensitivity due to their unfavorable geotechnical characteristics. Slope (0.47, p < 0.001) and distance from rivers (0.34, p < 0.01) were secondary influencing factors. This study demonstrated that integrating field methods and remote sensing with machine learning algorithms provides a powerful tool for landslide risk management. However, limitations include the resolution of input data and spatial scale of analysis. Simultaneous consideration of natural and human factors can make planning for high-risk areas more effective. Implementation of proposed solutions including biological measures, water resource control, and structural measures could lead to significant damage reduction. Future studies are recommended to use combined methods with higher resolution data and three-dimensional modeling approaches. This research aimed to map landslide susceptibility in the Dehsheikh watershed in northeastern Khuzestan province, identifying high-risk areas and proposing practical solutions using the Multiple Linear Regression (MLR) algorithm. The study examined 15 influential parameters across four main categories: geological factors (formations, faults, earthquakes), topographic factors (slope, aspect, elevation), environmental factors (precipitation, land use, vegetation cover), and hydrological factors (distance from rivers, Topographic Wetness Index (TWI), and SPI). In this study, 129 historical landslide points were identified and 2,500 random points were generated as control samples. The data were split into a 70:30 ratio for model training and testing. After preparing information layers in ArcGIS, SAGA-GIS, and ENVI software, modeling was performed using the MLR algorithm in the RStudio environment.<br /><br />The ROC curve validation showed the model had excellent performance in landslide prediction with AUC = 0.909 for training data and AUC = 0.906 for test data (p < 0.01). Factor analysis revealed the geological formations had the greatest impact on landslide occurrence (coefficient = 0.60, p < 0.001). Specifically, the Quaternary and Gachsaran formations showed the highest sensitivity due to their unfavorable geotechnical characteristics. Slope (0.47, p < 0.001) and distance from rivers (0.34, p < 0.01) were secondary influencing factors. This study demonstrated that integrating field methods and remote sensing with machine learning algorithms provides a powerful tool for landslide risk management. However, limitations include the resolution of input data and spatial scale of analysis. Simultaneous consideration of natural and human factors can make planning for high-risk areas more effective. Implementation of proposed solutions including biological measures, water resource control, and structural measures could lead to significant damage reduction. Future studies are recommended to use combined methods with higher resolution data and three-dimensional modeling approaches.